outlet/routing: read from RIB without any locks by vincentbernat · Pull Request #2244 · akvorado/akvorado

vincentbernat · 2026-02-17T14:35:25Z

This series modifies the BMP provider for the routing component by removing locks.

The intern package is replaced by Go's unique package. This is more efficient than trying to make the intern package thread-safe. Lookup is not impacted by the change, insertion is slower, but routes take less space and flushing is faster (compared to the thread-safe version of the intern package).

bart's persist tables are used to make the RIB thread-safe. This has a huge performance impact on insertion performance (×10), as well as space (×2). Impact on lookup is small (~ 7%).

The next step is to make paths thread-safe by using some kind of sync.Map. The last commit swith to xsync.Map which seems the most efficient for our use case.

At the end, 42% increase in space, 1000% increase in insertion time, 12% lookup. The benchmarks do not test performance of multiple routines updating/looking up the RIB, but the idea is that it should behave better without locks, even if it is slower during updates. So, hopefully, someone could test the result for real.

Fix #2092.

codecov · 2026-02-17T14:46:09Z

Codecov Report

❌ Patch coverage is 99.46237% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 84.93%. Comparing base (ed54b28) to head (3a736ab).
⚠️ Report is 28 commits behind head on main.

Files with missing lines	Patch %	Lines
outlet/routing/provider/bmp/events.go	98.18%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2244      +/-   ##
==========================================
+ Coverage   84.83%   84.93%   +0.09%     
==========================================
  Files         231      229       -2     
  Lines       14029    13909     -120     
==========================================
- Hits        11902    11813      -89     
+ Misses       1421     1397      -24     
+ Partials      706      699       -7

Flag	Coverage Δ
e2e	`42.91% <50.67%> (+0.02%)`	⬆️
unittests	`84.38% <99.46%> (+0.37%)`	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

When flushing RIBs or receiving a big update, some BMP receivers may not process the packets in a timely fashion and may be flagged as stuck, closing the connection. The workaround was to increase the kernel buffer. Instead, we separate BMP message processing into a goroutine for I/O processing and a goroutine for handling received messages, with a configurable buffered channels between them. Fix #2092 This may be better than #2244 as performance may be better (but no benchmark for that).

``` goos: linux goarch: amd64 pkg: akvorado/outlet/routing/provider/bmp cpu: Intel(R) Xeon(R) W-2295 CPU @ 3.00GHz │ bench-step0.txt │ bench-step1.txt │ │ bytes/route │ bytes/route vs base │ RIBInsertion/1000_routes,_1_peers-36 345.5 ± 4% 533.4 ± 3% +54.35% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 360.0 ± 3% 544.9 ± 4% +51.35% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 323.5 ± 0% 514.0 ± 0% +58.89% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 322.8 ± 0% 516.8 ± 0% +60.11% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 317.9 ± 0% 504.7 ± 0% +58.76% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 269.8 ± 0% 452.4 ± 0% +67.68% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 265.0 ± 0% 450.1 ± 0% +69.85% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 251.0 ± 0% 430.6 ± 0% +71.55% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 207.1 ± 1% 364.0 ± 0% +75.76% (p=0.002 n=6) geomean 291.8 475.5 +62.95% │ bench-step0.txt │ bench-step1.txt │ │ sec/route │ sec/route vs base │ RIBInsertion/1000_routes,_1_peers-36 1.135µ ± 9% 1.508µ ± 3% +32.82% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 1.133µ ± 5% 1.548µ ± 5% +36.64% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 1.112µ ± 5% 1.567µ ± 11% +40.87% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 1.189µ ± 8% 1.585µ ± 7% +33.32% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 1.242µ ± 12% 1.559µ ± 4% +25.57% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 1.205µ ± 7% 1.577µ ± 13% +30.93% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 1.086µ ± 8% 1.501µ ± 10% +38.28% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 1.037µ ± 7% 1.445µ ± 7% +39.41% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 1.154µ ± 6% 1.345µ ± 7% +16.56% (p=0.002 n=6) geomean 1.142µ 1.513µ +32.50% │ bench-step0.txt │ bench-step1.txt │ │ allocs/op │ allocs/op vs base │ RIBInsertion/1000_routes,_1_peers-36 2.069k ± 0% 4.804k ± 0% +132.19% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 3.376k ± 0% 9.066k ± 0% +168.54% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 6.900k ± 0% 21.722k ± 0% +214.81% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 13.23k ± 0% 43.45k ± 0% +228.46% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 27.52k ± 0% 87.29k ± 0% +217.21% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 81.70k ± 0% 228.76k ± 0% +180.01% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 183.0k ± 0% 475.3k ± 0% +159.80% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 354.7k ± 0% 918.2k ± 0% +158.87% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 690.1k ± 0% 1947.4k ± 0% +182.18% (p=0.002 n=6) RIBLookup/1000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/1000_routes,_1_peers-36 2.055k ± 0% 2.055k ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/1000_routes,_2_peers-36 3.363k ± 0% 3.363k ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/1000_routes,_5_peers-36 6.834k ± 0% 6.834k ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/10000_routes,_1_peers-36 13.01k ± 0% 13.01k ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/10000_routes,_2_peers-36 27.13k ± 0% 27.13k ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/10000_routes,_5_peers-36 81.30k ± 0% 81.30k ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/100000_routes,_1_peers-36 181.4k ± 0% 181.4k ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/100000_routes,_2_peers-36 352.1k ± 0% 352.1k ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/100000_routes,_5_peers-36 685.2k ± 0% 685.2k ± 0% ~ (p=1.000 n=6) ¹ RTAHash-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RTAEqual-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ geomean ² +33.70% ² ¹ all samples are equal ² summaries must be >0 to compute geomean │ bench-step0.txt │ bench-step1.txt │ │ sec/op │ sec/op vs base │ RIBLookup/1000_routes,_1_peers-36 30.39n ± 5% 29.78n ± 3% ~ (p=0.180 n=6) RIBLookup/1000_routes,_2_peers-36 30.30n ± 4% 30.51n ± 4% ~ (p=0.937 n=6) RIBLookup/1000_routes,_5_peers-36 32.59n ± 21% 31.74n ± 6% ~ (p=0.195 n=6) RIBLookup/10000_routes,_1_peers-36 33.58n ± 13% 33.85n ± 8% ~ (p=0.818 n=6) RIBLookup/10000_routes,_2_peers-36 37.09n ± 5% 35.28n ± 15% ~ (p=0.589 n=6) RIBLookup/10000_routes,_5_peers-36 44.42n ± 12% 40.30n ± 6% -9.29% (p=0.009 n=6) RIBLookup/100000_routes,_1_peers-36 88.08n ± 15% 87.19n ± 13% ~ (p=0.937 n=6) RIBLookup/100000_routes,_2_peers-36 96.64n ± 6% 95.36n ± 8% ~ (p=0.589 n=6) RIBLookup/100000_routes,_5_peers-36 102.50n ± 9% 96.85n ± 2% -5.51% (p=0.015 n=6) RIBLookup/1000000_routes,_1_peers-36 159.0n ± 8% 145.9n ± 22% ~ (p=0.065 n=6) RIBLookup/1000000_routes,_2_peers-36 157.9n ± 10% 142.4n ± 23% ~ (p=0.065 n=6) RIBLookup/1000000_routes,_5_peers-36 181.2n ± 17% 154.5n ± 8% -14.76% (p=0.004 n=6) RTAHash-36 19.30n ± 4% 18.68n ± 3% -3.21% (p=0.037 n=6) RTAEqual-36 8.700n ± 9% 6.985n ± 4% -19.71% (p=0.002 n=6) geomean 52.17n 49.06n -5.97% │ bench-step0.txt │ bench-step1.txt │ │ ms/op │ ms/op vs base │ RIBFlush/1000_routes,_1_peers-36 578.6m ± 8% 607.2m ± 2% +4.95% (p=0.015 n=6) RIBFlush/1000_routes,_2_peers-36 922.9m ± 8% 1023.5m ± 5% +10.90% (p=0.015 n=6) RIBFlush/1000_routes,_5_peers-36 1.825 ± 4% 1.949 ± 5% +6.77% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 5.619 ± 8% 6.077 ± 6% +8.15% (p=0.041 n=6) RIBFlush/10000_routes,_2_peers-36 9.527 ± 4% 10.735 ± 8% +12.68% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 20.67 ± 4% 20.75 ± 11% ~ (p=0.699 n=6) RIBFlush/100000_routes,_1_peers-36 76.62 ± 12% 69.47 ± 7% -9.33% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 114.3 ± 5% 119.2 ± 8% +4.29% (p=0.026 n=6) RIBFlush/100000_routes,_5_peers-36 218.8 ± 10% 221.4 ± 11% ~ (p=0.589 n=6) geomean 10.84 11.30 +4.25% ```

sync.Map does not have generic and therefore, there is a performance hit. Use a third-party map instead. ``` goos: linux goarch: amd64 pkg: akvorado/outlet/routing/provider/bmp cpu: Intel(R) Xeon(R) W-2295 CPU @ 3.00GHz │ bench-step0.txt │ bench-step3.txt │ │ bytes/route │ bytes/route vs base │ RIBInsertion/1000_routes,_1_peers-36 345.5 ± 4% 549.4 ± 2% +58.99% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 360.0 ± 3% 533.0 ± 2% +48.04% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 323.5 ± 0% 486.8 ± 0% +50.46% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 322.8 ± 0% 502.9 ± 0% +55.82% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 317.9 ± 0% 481.5 ± 0% +51.46% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 269.8 ± 0% 394.5 ± 0% +46.22% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 265.0 ± 0% 389.2 ± 0% +46.87% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 251.0 ± 0% 375.3 ± 0% +49.52% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 207.1 ± 1% 307.4 ± 0% +48.41% (p=0.002 n=6) geomean 291.8 439.5 +50.59% │ bench-step0.txt │ bench-step3.txt │ │ sec/route │ sec/route vs base │ RIBInsertion/1000_routes,_1_peers-36 1.135µ ± 9% 1.516µ ± 3% +33.57% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 1.133µ ± 5% 1.530µ ± 5% +35.06% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 1.112µ ± 5% 1.459µ ± 6% +31.21% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 1.189µ ± 8% 1.584µ ± 9% +33.28% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 1.242µ ± 12% 1.527µ ± 6% +22.96% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 1.205µ ± 7% 1.391µ ± 2% +15.48% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 1.086µ ± 8% 1.265µ ± 6% +16.54% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 1.037µ ± 7% 1.290µ ± 12% +24.46% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 1.154µ ± 6% 1.238µ ± 7% ~ (p=0.065 n=6) geomean 1.142µ 1.417µ +24.08% │ bench-step0.txt │ bench-step3.txt │ │ allocs/op │ allocs/op vs base │ RIBInsertion/1000_routes,_1_peers-36 2.069k ± 0% 2.556k ± 0% +23.54% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 3.376k ± 0% 3.927k ± 0% +16.32% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 6.900k ± 0% 7.547k ± 0% +9.38% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 13.23k ± 0% 13.94k ± 0% +5.40% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 27.52k ± 0% 28.30k ± 0% +2.83% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 81.70k ± 0% 82.54k ± 0% +1.03% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 183.0k ± 0% 183.9k ± 0% +0.50% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 354.7k ± 0% 355.6k ± 0% +0.27% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 690.1k ± 0% 691.1k ± 0% +0.15% (p=0.002 n=6) RIBLookup/1000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/1000_routes,_1_peers-36 2.055k ± 0% 2.055k ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/1000_routes,_2_peers-36 3.363k ± 0% 3.364k ± 0% +0.03% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 6.834k ± 0% 6.835k ± 0% +0.01% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 13.01k ± 0% 13.01k ± 0% -0.02% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 27.13k ± 0% 27.13k ± 0% -0.00% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 81.30k ± 0% 81.29k ± 0% -0.00% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 181.4k ± 0% 181.4k ± 0% -0.00% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 352.1k ± 0% 352.1k ± 0% -0.00% (p=0.002 n=6) RIBFlush/100000_routes,_5_peers-36 685.2k ± 0% 685.2k ± 0% -0.00% (p=0.002 n=6) RTAHash-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RTAEqual-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ geomean ² +1.74% ² ¹ all samples are equal ² summaries must be >0 to compute geomean │ bench-step0.txt │ bench-step3.txt │ │ sec/op │ sec/op vs base │ RIBLookup/1000_routes,_1_peers-36 30.39n ± 5% 29.55n ± 3% ~ (p=0.065 n=6) RIBLookup/1000_routes,_2_peers-36 30.30n ± 4% 29.53n ± 2% -2.56% (p=0.017 n=6) RIBLookup/1000_routes,_5_peers-36 32.59n ± 21% 30.55n ± 8% -6.26% (p=0.015 n=6) RIBLookup/10000_routes,_1_peers-36 33.58n ± 13% 33.19n ± 5% ~ (p=0.240 n=6) RIBLookup/10000_routes,_2_peers-36 37.09n ± 5% 38.77n ± 13% ~ (p=0.310 n=6) RIBLookup/10000_routes,_5_peers-36 44.42n ± 12% 41.77n ± 3% -5.97% (p=0.004 n=6) RIBLookup/100000_routes,_1_peers-36 88.08n ± 15% 85.46n ± 5% ~ (p=0.240 n=6) RIBLookup/100000_routes,_2_peers-36 96.64n ± 6% 96.53n ± 3% ~ (p=0.818 n=6) RIBLookup/100000_routes,_5_peers-36 102.50n ± 9% 97.91n ± 6% ~ (p=0.240 n=6) RIBLookup/1000000_routes,_1_peers-36 159.0n ± 8% 131.8n ± 13% -17.05% (p=0.002 n=6) RIBLookup/1000000_routes,_2_peers-36 157.9n ± 10% 141.5n ± 8% -10.41% (p=0.009 n=6) RIBLookup/1000000_routes,_5_peers-36 181.2n ± 17% 156.2n ± 3% -13.85% (p=0.009 n=6) RTAHash-36 19.30n ± 4% 18.80n ± 4% ~ (p=0.123 n=6) RTAEqual-36 8.700n ± 9% 6.913n ± 6% -20.54% (p=0.002 n=6) geomean 52.17n 48.83n -6.41% │ bench-step0.txt │ bench-step3.txt │ │ ms/op │ ms/op vs base │ RIBFlush/1000_routes,_1_peers-36 578.6m ± 8% 633.0m ± 10% +9.39% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 922.9m ± 8% 1083.0m ± 9% +17.35% (p=0.004 n=6) RIBFlush/1000_routes,_5_peers-36 1.825 ± 4% 1.917 ± 4% +5.04% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 5.619 ± 8% 5.899 ± 7% ~ (p=0.093 n=6) RIBFlush/10000_routes,_2_peers-36 9.527 ± 4% 10.405 ± 3% +9.22% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 20.67 ± 4% 20.09 ± 6% ~ (p=0.093 n=6) RIBFlush/100000_routes,_1_peers-36 76.62 ± 12% 66.96 ± 12% -12.61% (p=0.009 n=6) RIBFlush/100000_routes,_2_peers-36 114.3 ± 5% 112.5 ± 8% ~ (p=0.937 n=6) RIBFlush/100000_routes,_5_peers-36 218.8 ± 10% 219.8 ± 4% ~ (p=1.000 n=6) geomean 10.84 11.16 +2.94% ```

This is only a first step to compare performance of both. Next step would be to use bart's persist tables to remove more locking. ``` goos: linux goarch: amd64 pkg: akvorado/outlet/routing/provider/bmp cpu: Intel(R) Xeon(R) W-2295 CPU @ 3.00GHz │ bench-step0.txt │ bench-step4.txt │ │ bytes/route │ bytes/route vs base │ RIBInsertion/1000_routes,_1_peers-36 345.5 ± 4% 441.4 ± 8% +27.74% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 360.0 ± 3% 455.4 ± 7% +26.50% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 323.5 ± 0% 458.5 ± 0% +41.73% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 322.8 ± 0% 462.1 ± 0% +43.18% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 317.9 ± 0% 449.1 ± 0% +41.29% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 269.8 ± 0% 423.1 ± 0% +56.82% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 265.0 ± 0% 418.1 ± 0% +57.79% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 251.0 ± 0% 395.6 ± 0% +57.61% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 207.1 ± 1% 349.4 ± 0% +68.71% (p=0.002 n=6) geomean 291.8 426.6 +46.18% │ bench-step0.txt │ bench-step4.txt │ │ sec/route │ sec/route vs base │ RIBInsertion/1000_routes,_1_peers-36 1135.0n ± 9% 808.4n ± 3% -28.78% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 1132.5n ± 5% 854.7n ± 8% -24.53% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 1112.0n ± 5% 930.2n ± 10% -16.34% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 1.189µ ± 8% 1.752µ ± 11% +47.37% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 1.242µ ± 12% 2.576µ ± 14% +107.45% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 1.205µ ± 7% 1.959µ ± 35% +62.60% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 1.086µ ± 8% 3.137µ ± 9% +188.99% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 1.037µ ± 7% 2.904µ ± 11% +180.13% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 1.154µ ± 6% 2.557µ ± 8% +121.67% (p=0.002 n=6) geomean 1.142µ 1.721µ +50.71% │ bench-step0.txt │ bench-step4.txt │ │ allocs/op │ allocs/op vs base │ RIBInsertion/1000_routes,_1_peers-36 2.069k ± 0% 2.039k ± 1% -1.45% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 3.376k ± 0% 3.526k ± 2% +4.44% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 6.900k ± 0% 8.043k ± 7% +16.56% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 13.23k ± 0% 34.11k ± 19% +157.83% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 27.52k ± 0% 93.81k ± 14% +240.91% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 81.70k ± 0% 181.67k ± 58% +122.37% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 183.0k ± 0% 866.0k ± 0% +373.33% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 354.7k ± 0% 1673.8k ± 0% +371.94% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 690.1k ± 0% 3688.4k ± 0% +434.47% (p=0.002 n=6) RIBLookup/1000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/1000_routes,_1_peers-36 2.055k ± 0% 2.042k ± 0% -0.63% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 3.363k ± 0% 3.351k ± 0% -0.36% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 6.834k ± 0% 6.822k ± 0% -0.18% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 13.01k ± 0% 12.99k ± 0% -0.14% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 27.13k ± 0% 27.12k ± 0% -0.06% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 81.30k ± 0% 81.28k ± 0% -0.02% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 181.4k ± 0% 181.4k ± 0% -0.01% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 352.1k ± 0% 352.1k ± 0% -0.01% (p=0.002 n=6) RIBFlush/100000_routes,_5_peers-36 685.2k ± 0% 685.1k ± 0% -0.00% (p=0.002 n=6) RTAHash-36 0.000 ± 0% RTAEqual-36 0.000 ± 0% geomean ² +30.23% ³ ² ¹ all samples are equal ² summaries must be >0 to compute geomean ³ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step4.txt │ │ sec/op │ sec/op vs base │ RIBLookup/1000_routes,_1_peers-36 30.39n ± 5% 30.30n ± 4% ~ (p=0.818 n=6) RIBLookup/1000_routes,_2_peers-36 30.30n ± 4% 30.50n ± 23% ~ (p=0.937 n=6) RIBLookup/1000_routes,_5_peers-36 32.59n ± 21% 31.24n ± 3% -4.13% (p=0.009 n=6) RIBLookup/10000_routes,_1_peers-36 33.58n ± 13% 34.85n ± 5% ~ (p=0.589 n=6) RIBLookup/10000_routes,_2_peers-36 37.09n ± 5% 35.53n ± 3% ~ (p=0.093 n=6) RIBLookup/10000_routes,_5_peers-36 44.42n ± 12% 39.40n ± 5% -11.32% (p=0.002 n=6) RIBLookup/100000_routes,_1_peers-36 88.08n ± 15% 93.84n ± 2% ~ (p=0.145 n=6) RIBLookup/100000_routes,_2_peers-36 96.64n ± 6% 103.10n ± 8% +6.68% (p=0.041 n=6) RIBLookup/100000_routes,_5_peers-36 102.5n ± 9% 109.2n ± 7% ~ (p=0.132 n=6) RIBLookup/1000000_routes,_1_peers-36 159.0n ± 8% 134.2n ± 13% -15.54% (p=0.002 n=6) RIBLookup/1000000_routes,_2_peers-36 157.9n ± 10% 146.1n ± 8% -7.47% (p=0.009 n=6) RIBLookup/1000000_routes,_5_peers-36 181.2n ± 17% 168.2n ± 12% ~ (p=0.093 n=6) RTAHash-36 19.30n ± 4% RTAEqual-36 8.700n ± 9% geomean 52.17n 64.21n -2.43% ¹ ¹ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step4.txt │ │ ms/op │ ms/op vs base │ RIBFlush/1000_routes,_1_peers-36 578.6m ± 8% 360.7m ± 7% -37.67% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 922.9m ± 8% 650.1m ± 4% -29.56% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 1.825 ± 4% 1.375 ± 7% -24.71% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 5.619 ± 8% 2.708 ± 5% -51.79% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 9.527 ± 4% 5.314 ± 4% -44.22% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 20.67 ± 4% 16.62 ± 8% -19.59% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 76.62 ± 12% 44.00 ± 9% -42.58% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 114.35 ± 5% 88.05 ± 6% -23.00% (p=0.002 n=6) RIBFlush/100000_routes,_5_peers-36 218.8 ± 10% 199.9 ± 7% -8.62% (p=0.015 n=6) geomean 10.84 7.312 -32.55% ```

Remove global locking when using the RIB tree by using bart's persist tables and some atomic pointers. Lock is not used anymore when reading. ``` goos: linux goarch: amd64 pkg: akvorado/outlet/routing/provider/bmp cpu: Intel(R) Xeon(R) W-2295 CPU @ 3.00GHz │ bench-step0.txt │ bench-step5.txt │ │ bytes/route │ bytes/route vs base │ RIBInsertion/1000_routes,_1_peers-36 345.5 ± 4% 437.7 ± 8% +26.67% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 360.0 ± 3% 448.4 ± 6% +24.57% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 323.5 ± 0% 451.9 ± 0% +39.68% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 322.8 ± 0% 450.8 ± 0% +39.67% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 317.9 ± 0% 441.8 ± 0% +38.97% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 269.8 ± 0% 415.7 ± 0% +54.08% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 265.0 ± 0% 408.2 ± 0% +54.04% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 251.0 ± 0% 386.7 ± 0% +54.06% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 207.1 ± 1% 340.1 ± 0% +64.22% (p=0.002 n=6) geomean 291.8 418.6 +43.43% │ bench-step0.txt │ bench-step5.txt │ │ sec/route │ sec/route vs base │ RIBInsertion/1000_routes,_1_peers-36 1.135µ ± 9% 10.229µ ± 7% +801.19% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 1.133µ ± 5% 9.068µ ± 5% +700.66% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 1.112µ ± 5% 9.625µ ± 4% +765.56% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 1.189µ ± 8% 11.386µ ± 7% +858.06% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 1.242µ ± 12% 12.474µ ± 10% +904.75% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 1.205µ ± 7% 15.233µ ± 9% +1164.67% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 1.086µ ± 8% 15.196µ ± 6% +1299.91% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 1.037µ ± 7% 14.242µ ± 5% +1274.00% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 1.154µ ± 6% 11.111µ ± 7% +863.20% (p=0.002 n=6) geomean 1.142µ 11.86µ +938.80% │ bench-step0.txt │ bench-step5.txt │ │ allocs/op │ allocs/op vs base │ RIBInsertion/1000_routes,_1_peers-36 2.069k ± 0% 57.416k ± 0% +2675.08% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 3.376k ± 0% 76.677k ± 0% +2171.24% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 6.900k ± 0% 150.166k ± 1% +2076.32% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 13.23k ± 0% 340.04k ± 1% +2470.38% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 27.52k ± 0% 902.70k ± 0% +3180.38% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 81.70k ± 0% 3486.34k ± 0% +4167.41% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 183.0k ± 0% 8419.7k ± 0% +4502.17% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 354.7k ± 0% 15275.3k ± 0% +4206.86% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 690.1k ± 0% 23215.0k ± 0% +3263.96% (p=0.002 n=6) RIBLookup/1000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/1000_routes,_1_peers-36 2055.00 ± 0% 32.00 ± 0% -98.44% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 3.363k ± 0% 2.046k ± 0% -39.16% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 6.834k ± 0% 5.650k ± 0% -17.33% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 13011.00 ± 0% 97.00 ± 0% -99.25% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 27.13k ± 0% 12.97k ± 0% -52.18% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 81.30k ± 0% 62.18k ± 0% -23.51% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 181413.0 ± 0% 557.0 ± 0% -99.69% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 352.1k ± 0% 181.8k ± 0% -48.37% (p=0.002 n=6) RIBFlush/100000_routes,_5_peers-36 685.2k ± 0% 590.5k ± 0% -13.81% (p=0.002 n=6) RTAHash-36 0.000 ± 0% RTAEqual-36 0.000 ± 0% geomean ² +58.28% ³ ² ¹ all samples are equal ² summaries must be >0 to compute geomean ³ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step5.txt │ │ sec/op │ sec/op vs base │ RIBLookup/1000_routes,_1_peers-36 30.39n ± 5% 29.45n ± 2% -3.11% (p=0.009 n=6) RIBLookup/1000_routes,_2_peers-36 30.30n ± 4% 30.03n ± 6% ~ (p=0.589 n=6) RIBLookup/1000_routes,_5_peers-36 32.59n ± 21% 31.57n ± 2% -3.15% (p=0.041 n=6) RIBLookup/10000_routes,_1_peers-36 33.58n ± 13% 35.15n ± 4% ~ (p=0.180 n=6) RIBLookup/10000_routes,_2_peers-36 37.09n ± 5% 37.06n ± 3% ~ (p=0.937 n=6) RIBLookup/10000_routes,_5_peers-36 44.42n ± 12% 43.22n ± 2% ~ (p=0.310 n=6) RIBLookup/100000_routes,_1_peers-36 88.08n ± 15% 99.44n ± 3% ~ (p=0.065 n=6) RIBLookup/100000_routes,_2_peers-36 96.64n ± 6% 110.70n ± 3% +14.55% (p=0.002 n=6) RIBLookup/100000_routes,_5_peers-36 102.5n ± 9% 118.6n ± 1% +15.71% (p=0.002 n=6) RIBLookup/1000000_routes,_1_peers-36 159.0n ± 8% 173.3n ± 13% ~ (p=0.126 n=6) RIBLookup/1000000_routes,_2_peers-36 157.9n ± 10% 194.1n ± 16% +22.86% (p=0.009 n=6) RIBLookup/1000000_routes,_5_peers-36 181.2n ± 17% 233.1n ± 21% +28.63% (p=0.015 n=6) RTAHash-36 19.30n ± 4% RTAEqual-36 8.700n ± 9% geomean 52.17n 70.89n +7.72% ¹ ¹ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step5.txt │ │ ms/op │ ms/op vs base │ RIBFlush/1000_routes,_1_peers-36 578.6m ± 8% 515.0m ± 2% -10.99% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 922.9m ± 8% 1131.0m ± 3% +22.55% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 1.825 ± 4% 2.067 ± 1% +13.20% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 5.619 ± 8% 4.821 ± 3% -14.19% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 9.527 ± 4% 9.193 ± 4% ~ (p=0.065 n=6) RIBFlush/10000_routes,_5_peers-36 20.67 ± 4% 25.31 ± 13% +22.42% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 76.62 ± 12% 48.98 ± 5% -36.08% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 114.35 ± 5% 99.05 ± 8% -13.38% (p=0.002 n=6) RIBFlush/100000_routes,_5_peers-36 218.8 ± 10% 203.6 ± 7% -6.95% (p=0.015 n=6) geomean 10.84 10.32 -4.76% ```

We only keep a lock when writing (mostly to find a free slot and to modifies peers). Readers are not using locks. ``` goos: linux goarch: amd64 pkg: akvorado/outlet/routing/provider/bmp cpu: Intel(R) Xeon(R) W-2295 CPU @ 3.00GHz │ bench-step0.txt │ bench-step6.txt │ │ bytes/route │ bytes/route vs base │ RIBInsertion/1000_routes,_1_peers-36 345.5 ± 4% 531.8 ± 5% +53.90% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 360.0 ± 3% 508.1 ± 1% +41.13% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 323.5 ± 0% 513.8 ± 0% +58.81% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 322.8 ± 0% 516.2 ± 0% +59.94% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 317.9 ± 0% 501.1 ± 0% +57.64% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 269.8 ± 0% 490.8 ± 0% +81.89% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 265.0 ± 0% 490.6 ± 0% +85.15% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 251.0 ± 0% 469.9 ± 0% +87.19% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 207.1 ± 1% 415.6 ± 0% +100.68% (p=0.002 n=6) geomean 291.8 492.0 +68.59% │ bench-step0.txt │ bench-step6.txt │ │ sec/route │ sec/route vs base │ RIBInsertion/1000_routes,_1_peers-36 1.135µ ± 9% 10.067µ ± 6% +786.92% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 1.133µ ± 5% 9.056µ ± 4% +699.60% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 1.112µ ± 5% 9.803µ ± 4% +781.56% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 1.189µ ± 8% 11.467µ ± 10% +864.79% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 1.242µ ± 12% 12.938µ ± 13% +942.13% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 1.205µ ± 7% 15.343µ ± 4% +1173.81% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 1.086µ ± 8% 14.382µ ± 11% +1224.92% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 1.037µ ± 7% 13.902µ ± 3% +1241.24% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 1.154µ ± 6% 11.242µ ± 4% +874.56% (p=0.002 n=6) geomean 1.142µ 11.84µ +936.99% │ bench-step0.txt │ bench-step6.txt │ │ allocs/op │ allocs/op vs base │ RIBInsertion/1000_routes,_1_peers-36 2.069k ± 0% 60.337k ± 0% +2816.24% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 3.376k ± 0% 82.416k ± 0% +2341.23% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 6.900k ± 0% 165.111k ± 0% +2292.91% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 13.23k ± 0% 368.91k ± 2% +2688.63% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 27.52k ± 0% 959.28k ± 1% +3386.02% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 81.70k ± 0% 3651.68k ± 0% +4369.79% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 183.0k ± 0% 8724.2k ± 0% +4668.61% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 354.7k ± 0% 15883.7k ± 0% +4378.40% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 690.1k ± 0% 24711.9k ± 0% +3480.87% (p=0.002 n=6) RIBLookup/1000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/1000_routes,_1_peers-36 2055.00 ± 0% 32.00 ± 0% -98.44% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 3.363k ± 0% 2.049k ± 0% -39.07% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 6.834k ± 0% 5.653k ± 0% -17.28% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 13011.00 ± 0% 97.00 ± 0% -99.25% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 27.13k ± 0% 12.99k ± 0% -52.14% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 81.30k ± 0% 62.22k ± 0% -23.47% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 181413.0 ± 0% 557.0 ± 0% -99.69% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 352.1k ± 0% 182.8k ± 0% -48.08% (p=0.002 n=6) RIBFlush/100000_routes,_5_peers-36 685.2k ± 0% 594.4k ± 0% -13.24% (p=0.002 n=6) RTAHash-36 0.000 ± 0% RTAEqual-36 0.000 ± 0% geomean ² +61.25% ³ ² ¹ all samples are equal ² summaries must be >0 to compute geomean ³ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step6.txt │ │ sec/op │ sec/op vs base │ RIBLookup/1000_routes,_1_peers-36 30.39n ± 5% 31.07n ± 7% ~ (p=0.240 n=6) RIBLookup/1000_routes,_2_peers-36 30.30n ± 4% 30.91n ± 3% ~ (p=0.180 n=6) RIBLookup/1000_routes,_5_peers-36 32.59n ± 21% 32.56n ± 5% ~ (p=0.937 n=6) RIBLookup/10000_routes,_1_peers-36 33.58n ± 13% 34.82n ± 4% ~ (p=0.310 n=6) RIBLookup/10000_routes,_2_peers-36 37.09n ± 5% 37.34n ± 2% ~ (p=0.818 n=6) RIBLookup/10000_routes,_5_peers-36 44.42n ± 12% 44.30n ± 4% ~ (p=1.000 n=6) RIBLookup/100000_routes,_1_peers-36 88.08n ± 15% 99.11n ± 7% ~ (p=0.093 n=6) RIBLookup/100000_routes,_2_peers-36 96.64n ± 6% 112.75n ± 8% +16.67% (p=0.002 n=6) RIBLookup/100000_routes,_5_peers-36 102.5n ± 9% 122.2n ± 3% +19.22% (p=0.002 n=6) RIBLookup/1000000_routes,_1_peers-36 159.0n ± 8% 172.9n ± 15% ~ (p=0.132 n=6) RIBLookup/1000000_routes,_2_peers-36 157.9n ± 10% 198.8n ± 21% +25.83% (p=0.026 n=6) RIBLookup/1000000_routes,_5_peers-36 181.2n ± 17% 232.2n ± 20% +28.11% (p=0.004 n=6) RTAHash-36 19.30n ± 4% RTAEqual-36 8.700n ± 9% geomean 52.17n 72.07n +9.51% ¹ ¹ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step6.txt │ │ ms/op │ ms/op vs base │ RIBFlush/1000_routes,_1_peers-36 578.6m ± 8% 774.4m ± 1% +33.83% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 922.9m ± 8% 1522.0m ± 3% +64.91% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 1.825 ± 4% 3.127 ± 3% +71.27% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 5.619 ± 8% 7.637 ± 3% +35.93% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 9.527 ± 4% 14.060 ± 5% +47.58% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 20.67 ± 4% 35.38 ± 9% +71.12% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 76.62 ± 12% 75.82 ± 5% ~ (p=0.589 n=6) RIBFlush/100000_routes,_2_peers-36 114.3 ± 5% 158.3 ± 15% +38.43% (p=0.002 n=6) RIBFlush/100000_routes,_5_peers-36 218.8 ± 10% 373.8 ± 3% +70.84% (p=0.002 n=6) geomean 10.84 15.84 +46.12% ```

sync.Map is using "any" and therefore, there is a performance hit. Instead, use xsync.Map that is mostly a dropin replacement without this issue. I did not bench it against concurrent swiss maps that were tested in an earlier commit. The benchmarks are a bit artificial, I don't think this is worth it. ``` goos: linux goarch: amd64 pkg: akvorado/outlet/routing/provider/bmp cpu: Intel(R) Xeon(R) W-2295 CPU @ 3.00GHz │ bench-step0.txt │ bench-step7.txt │ │ bytes/route │ bytes/route vs base │ RIBInsertion/1000_routes,_1_peers-36 345.5 ± 4% 452.4 ± 7% +30.92% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 360.0 ± 3% 433.2 ± 0% +20.33% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 323.5 ± 0% 441.1 ± 0% +36.37% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 322.8 ± 0% 441.4 ± 0% +36.76% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 317.9 ± 0% 432.9 ± 0% +36.17% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 269.8 ± 0% 418.6 ± 0% +55.17% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 265.0 ± 0% 411.2 ± 0% +55.17% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 251.0 ± 0% 389.4 ± 0% +55.14% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 207.1 ± 1% 338.4 ± 0% +63.42% (p=0.002 n=6) geomean 291.8 416.2 +42.63% │ bench-step0.txt │ bench-step7.txt │ │ sec/route │ sec/route vs base │ RIBInsertion/1000_routes,_1_peers-36 1.135µ ± 9% 9.984µ ± 4% +779.65% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 1.133µ ± 5% 8.827µ ± 4% +679.38% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 1.112µ ± 5% 9.790µ ± 9% +780.35% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 1.189µ ± 8% 11.927µ ± 6% +903.49% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 1.242µ ± 12% 12.903µ ± 10% +939.27% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 1.205µ ± 7% 15.608µ ± 12% +1195.81% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 1.086µ ± 8% 15.154µ ± 9% +1296.04% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 1.037µ ± 7% 14.102µ ± 4% +1260.54% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 1.154µ ± 6% 11.469µ ± 5% +894.23% (p=0.002 n=6) geomean 1.142µ 11.98µ +949.16% │ bench-step0.txt │ bench-step7.txt │ │ allocs/op │ allocs/op vs base │ RIBInsertion/1000_routes,_1_peers-36 2.069k ± 0% 58.341k ± 0% +2719.77% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 3.376k ± 0% 78.516k ± 1% +2225.70% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 6.900k ± 0% 154.131k ± 1% +2133.79% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 13.23k ± 0% 351.93k ± 1% +2560.31% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 27.52k ± 0% 920.91k ± 0% +3246.56% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 81.70k ± 0% 3548.12k ± 1% +4243.02% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 183.0k ± 0% 8517.4k ± 0% +4555.57% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 354.7k ± 0% 15470.1k ± 0% +4261.78% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 690.1k ± 0% 23703.6k ± 0% +3334.77% (p=0.002 n=6) RIBLookup/1000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/1000_routes,_1_peers-36 2055.00 ± 0% 41.00 ± 0% -98.00% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 3.363k ± 0% 2.047k ± 0% -39.13% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 6.834k ± 0% 5.651k ± 0% -17.31% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 13011.0 ± 0% 118.0 ± 0% -99.09% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 27.13k ± 0% 12.98k ± 0% -52.17% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 81.30k ± 0% 62.19k ± 0% -23.50% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 181413.0 ± 0% 587.0 ± 0% -99.68% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 352.1k ± 0% 182.1k ± 0% -48.28% (p=0.002 n=6) RIBFlush/100000_routes,_5_peers-36 685.2k ± 0% 591.8k ± 0% -13.62% (p=0.002 n=6) RTAHash-36 0.000 ± 0% RTAEqual-36 0.000 ± 0% geomean ² +61.93% ³ ² ¹ all samples are equal ² summaries must be >0 to compute geomean ³ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step7.txt │ │ sec/op │ sec/op vs base │ RIBLookup/1000_routes,_1_peers-36 30.39n ± 5% 30.03n ± 2% ~ (p=0.563 n=6) RIBLookup/1000_routes,_2_peers-36 30.30n ± 4% 30.83n ± 4% ~ (p=0.180 n=6) RIBLookup/1000_routes,_5_peers-36 32.59n ± 21% 32.25n ± 9% ~ (p=0.818 n=6) RIBLookup/10000_routes,_1_peers-36 33.58n ± 13% 34.56n ± 3% ~ (p=0.310 n=6) RIBLookup/10000_routes,_2_peers-36 37.09n ± 5% 36.45n ± 3% ~ (p=0.589 n=6) RIBLookup/10000_routes,_5_peers-36 44.42n ± 12% 43.58n ± 9% ~ (p=0.937 n=6) RIBLookup/100000_routes,_1_peers-36 88.08n ± 15% 96.90n ± 7% +10.01% (p=0.026 n=6) RIBLookup/100000_routes,_2_peers-36 96.64n ± 6% 111.05n ± 2% +14.91% (p=0.002 n=6) RIBLookup/100000_routes,_5_peers-36 102.5n ± 9% 120.0n ± 4% +17.07% (p=0.002 n=6) RIBLookup/1000000_routes,_1_peers-36 159.0n ± 8% 170.1n ± 12% ~ (p=0.221 n=6) RIBLookup/1000000_routes,_2_peers-36 157.9n ± 10% 196.3n ± 12% +24.25% (p=0.002 n=6) RIBLookup/1000000_routes,_5_peers-36 181.2n ± 17% 228.9n ± 19% +26.26% (p=0.015 n=6) RTAHash-36 19.30n ± 4% RTAEqual-36 8.700n ± 9% geomean 52.17n 70.91n +7.76% ¹ ¹ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step7.txt │ │ ms/op │ ms/op vs base │ RIBFlush/1000_routes,_1_peers-36 578.6m ± 8% 737.3m ± 1% +27.44% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 922.9m ± 8% 1373.5m ± 3% +48.82% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 1.825 ± 4% 2.542 ± 3% +39.22% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 5.619 ± 8% 6.957 ± 2% +23.82% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 9.527 ± 4% 10.670 ± 4% +12.00% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 20.67 ± 4% 27.73 ± 9% +34.15% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 76.62 ± 12% 58.81 ± 8% -23.25% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 114.3 ± 5% 105.5 ± 5% -7.78% (p=0.009 n=6) RIBFlush/100000_routes,_5_peers-36 218.8 ± 10% 227.0 ± 6% ~ (p=0.485 n=6) geomean 10.84 12.50 +15.32% ```

They are not modified anymore and can be stored directly. ``` goos: linux goarch: amd64 pkg: akvorado/outlet/routing/provider/bmp cpu: Intel(R) Xeon(R) W-2295 CPU @ 3.00GHz │ bench-step0.txt │ bench-step8.txt │ │ bytes/route │ bytes/route vs base │ RIBInsertion/1000_routes,_1_peers-36 345.5 ± 4% 454.5 ± 7% +31.53% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 360.0 ± 3% 433.6 ± 0% +20.46% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 323.5 ± 0% 442.3 ± 1% +36.72% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 322.8 ± 0% 442.0 ± 0% +36.95% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 317.9 ± 0% 433.1 ± 0% +36.22% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 269.8 ± 0% 418.6 ± 0% +55.15% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 265.0 ± 0% 411.3 ± 0% +55.21% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 251.0 ± 0% 389.4 ± 0% +55.14% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 207.1 ± 1% 338.6 ± 0% +63.50% (p=0.002 n=6) geomean 291.8 416.7 +42.79% │ bench-step0.txt │ bench-step8.txt │ │ sec/route │ sec/route vs base │ RIBInsertion/1000_routes,_1_peers-36 1.135µ ± 9% 10.123µ ± 4% +791.85% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 1.133µ ± 5% 8.864µ ± 6% +682.74% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 1.112µ ± 5% 9.658µ ± 6% +768.48% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 1.189µ ± 8% 11.451µ ± 6% +863.48% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 1.242µ ± 12% 13.130µ ± 2% +957.55% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 1.205µ ± 7% 16.346µ ± 7% +1257.08% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 1.086µ ± 8% 15.022µ ± 5% +1283.83% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 1.037µ ± 7% 14.529µ ± 7% +1301.74% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 1.154µ ± 6% 11.591µ ± 5% +904.85% (p=0.002 n=6) geomean 1.142µ 12.06µ +956.08% │ bench-step0.txt │ bench-step8.txt │ │ allocs/op │ allocs/op vs base │ RIBInsertion/1000_routes,_1_peers-36 2.069k ± 0% 58.406k ± 0% +2722.89% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 3.376k ± 0% 78.456k ± 0% +2223.92% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 6.900k ± 0% 154.810k ± 1% +2143.62% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 13.23k ± 0% 349.60k ± 1% +2542.69% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 27.52k ± 0% 921.70k ± 0% +3249.45% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 81.70k ± 0% 3545.51k ± 0% +4239.82% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 183.0k ± 0% 8517.4k ± 0% +4555.55% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 354.7k ± 0% 15470.1k ± 0% +4261.78% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 690.1k ± 0% 23703.9k ± 0% +3334.80% (p=0.002 n=6) RIBLookup/1000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/1000_routes,_1_peers-36 2055.00 ± 0% 41.00 ± 0% -98.00% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 3.363k ± 0% 2.047k ± 0% -39.13% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 6.834k ± 0% 5.651k ± 0% -17.31% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 13011.0 ± 0% 118.0 ± 0% -99.09% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 27.13k ± 0% 12.98k ± 0% -52.17% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 81.30k ± 0% 62.19k ± 0% -23.50% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 181413.0 ± 0% 587.0 ± 0% -99.68% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 352.1k ± 0% 182.1k ± 0% -48.28% (p=0.002 n=6) RIBFlush/100000_routes,_5_peers-36 685.2k ± 0% 591.8k ± 0% -13.62% (p=0.002 n=6) RTAHash-36 0.000 ± 0% RTAEqual-36 0.000 ± 0% geomean ² +61.92% ³ ² ¹ all samples are equal ² summaries must be >0 to compute geomean ³ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step8.txt │ │ sec/op │ sec/op vs base │ RIBLookup/1000_routes,_1_peers-36 30.39n ± 5% 30.32n ± 3% ~ (p=0.589 n=6) RIBLookup/1000_routes,_2_peers-36 30.30n ± 4% 29.72n ± 21% ~ (p=0.180 n=6) RIBLookup/1000_routes,_5_peers-36 32.59n ± 21% 31.79n ± 3% ~ (p=0.193 n=6) RIBLookup/10000_routes,_1_peers-36 33.58n ± 13% 34.92n ± 5% ~ (p=0.240 n=6) RIBLookup/10000_routes,_2_peers-36 37.09n ± 5% 38.45n ± 5% ~ (p=0.132 n=6) RIBLookup/10000_routes,_5_peers-36 44.42n ± 12% 44.34n ± 5% ~ (p=0.818 n=6) RIBLookup/100000_routes,_1_peers-36 88.08n ± 15% 99.03n ± 4% ~ (p=0.065 n=6) RIBLookup/100000_routes,_2_peers-36 96.64n ± 6% 112.60n ± 6% +16.51% (p=0.002 n=6) RIBLookup/100000_routes,_5_peers-36 102.5n ± 9% 119.6n ± 3% +16.68% (p=0.002 n=6) RIBLookup/1000000_routes,_1_peers-36 159.0n ± 8% 174.7n ± 17% ~ (p=0.065 n=6) RIBLookup/1000000_routes,_2_peers-36 157.9n ± 10% 195.8n ± 15% +23.99% (p=0.002 n=6) RIBLookup/1000000_routes,_5_peers-36 181.2n ± 17% 230.0n ± 18% +26.90% (p=0.009 n=6) RTAHash-36 19.30n ± 4% RTAEqual-36 8.700n ± 9% geomean 52.17n 71.51n +8.67% ¹ ¹ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step8.txt │ │ ms/op │ ms/op vs base │ RIBFlush/1000_routes,_1_peers-36 578.6m ± 8% 739.0m ± 1% +27.73% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 922.9m ± 8% 1355.0m ± 2% +46.82% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 1.825 ± 4% 2.593 ± 4% +42.02% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 5.619 ± 8% 7.162 ± 6% +27.47% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 9.527 ± 4% 11.260 ± 3% +18.19% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 20.67 ± 4% 29.70 ± 6% +43.65% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 76.62 ± 12% 63.95 ± 6% -16.55% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 114.3 ± 5% 117.8 ± 14% ~ (p=0.180 n=6) RIBFlush/100000_routes,_5_peers-36 218.8 ± 10% 254.0 ± 8% +16.11% (p=0.015 n=6) geomean 10.84 13.17 +21.47% ```

``` goos: linux goarch: amd64 pkg: akvorado/outlet/routing/provider/bmp cpu: Intel(R) Xeon(R) W-2295 CPU @ 3.00GHz │ bench-step0.txt │ bench-step9.txt │ │ bytes/route │ bytes/route vs base │ RIBInsertion/1000_routes,_1_peers-36 345.5 ± 4% 453.4 ± 7% +31.23% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 360.0 ± 3% 432.4 ± 1% +20.12% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 323.5 ± 0% 441.6 ± 1% +36.51% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 322.8 ± 0% 441.2 ± 0% +36.72% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 317.9 ± 0% 432.3 ± 0% +35.99% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 269.8 ± 0% 418.6 ± 0% +55.15% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 265.0 ± 0% 411.4 ± 0% +55.25% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 251.0 ± 0% 389.5 ± 0% +55.18% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 207.1 ± 1% 338.5 ± 0% +63.45% (p=0.002 n=6) geomean 291.8 416.2 +42.64% │ bench-step0.txt │ bench-step9.txt │ │ sec/route │ sec/route vs base │ RIBInsertion/1000_routes,_1_peers-36 1.135µ ± 9% 10.434µ ± 4% +819.25% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 1.133µ ± 5% 8.652µ ± 3% +663.97% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 1.112µ ± 5% 9.967µ ± 7% +796.31% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 1.189µ ± 8% 12.380µ ± 14% +941.65% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 1.242µ ± 12% 13.539µ ± 13% +990.50% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 1.205µ ± 7% 15.933µ ± 6% +1222.75% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 1.086µ ± 8% 15.746µ ± 14% +1350.58% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 1.037µ ± 7% 14.591µ ± 9% +1307.67% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 1.154µ ± 6% 11.803µ ± 4% +923.19% (p=0.002 n=6) geomean 1.142µ 12.31µ +978.61% │ bench-step0.txt │ bench-step9.txt │ │ allocs/op │ allocs/op vs base │ RIBInsertion/1000_routes,_1_peers-36 2.069k ± 0% 58.440k ± 0% +2724.55% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 3.376k ± 0% 78.376k ± 1% +2221.55% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 6.900k ± 0% 155.882k ± 1% +2159.16% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 13.23k ± 0% 351.68k ± 1% +2558.43% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 27.52k ± 0% 921.89k ± 0% +3250.15% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 81.70k ± 0% 3553.72k ± 0% +4249.87% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 183.0k ± 0% 8533.5k ± 0% +4564.36% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 354.7k ± 0% 15520.3k ± 0% +4275.95% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 690.1k ± 0% 23824.0k ± 0% +3352.21% (p=0.002 n=6) RIBLookup/1000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/1000_routes,_1_peers-36 2055.00 ± 0% 41.00 ± 0% -98.00% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 3.363k ± 0% 2.047k ± 0% -39.13% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 6.834k ± 0% 5.651k ± 0% -17.31% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 13011.0 ± 0% 118.0 ± 0% -99.09% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 27.13k ± 0% 12.98k ± 0% -52.17% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 81.30k ± 0% 62.19k ± 0% -23.50% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 181413.0 ± 0% 587.0 ± 0% -99.68% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 352.1k ± 0% 182.1k ± 0% -48.28% (p=0.002 n=6) RIBFlush/100000_routes,_5_peers-36 685.2k ± 0% 591.8k ± 0% -13.62% (p=0.002 n=6) RTAHash-36 0.000 ± 0% RTAEqual-36 0.000 ± 0% geomean ² +62.05% ³ ² ¹ all samples are equal ² summaries must be >0 to compute geomean ³ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step9.txt │ │ sec/op │ sec/op vs base │ RIBLookup/1000_routes,_1_peers-36 30.39n ± 5% 29.83n ± 4% ~ (p=0.310 n=6) RIBLookup/1000_routes,_2_peers-36 30.30n ± 4% 32.22n ± 3% +6.34% (p=0.004 n=6) RIBLookup/1000_routes,_5_peers-36 32.59n ± 21% 33.41n ± 4% ~ (p=0.699 n=6) RIBLookup/10000_routes,_1_peers-36 33.58n ± 13% 35.77n ± 4% ~ (p=0.093 n=6) RIBLookup/10000_routes,_2_peers-36 37.09n ± 5% 38.65n ± 6% +4.19% (p=0.009 n=6) RIBLookup/10000_routes,_5_peers-36 44.42n ± 12% 45.89n ± 7% ~ (p=0.485 n=6) RIBLookup/100000_routes,_1_peers-36 88.08n ± 15% 103.34n ± 9% +17.33% (p=0.015 n=6) RIBLookup/100000_routes,_2_peers-36 96.64n ± 6% 113.75n ± 8% +17.70% (p=0.002 n=6) RIBLookup/100000_routes,_5_peers-36 102.5n ± 9% 121.5n ± 9% +18.54% (p=0.004 n=6) RIBLookup/1000000_routes,_1_peers-36 159.0n ± 8% 171.6n ± 9% +7.96% (p=0.026 n=6) RIBLookup/1000000_routes,_2_peers-36 157.9n ± 10% 200.3n ± 17% +26.84% (p=0.004 n=6) RIBLookup/1000000_routes,_5_peers-36 181.2n ± 17% 227.6n ± 15% +25.57% (p=0.009 n=6) RTAHash-36 19.30n ± 4% RTAEqual-36 8.700n ± 9% geomean 52.17n 72.97n +10.88% ¹ ¹ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step9.txt │ │ ms/op │ ms/op vs base │ RIBFlush/1000_routes,_1_peers-36 578.6m ± 8% 726.5m ± 2% +25.55% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 922.9m ± 8% 1349.0m ± 2% +46.17% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 1.825 ± 4% 2.498 ± 3% +36.84% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 5.619 ± 8% 7.110 ± 4% +26.55% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 9.527 ± 4% 10.705 ± 4% +12.36% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 20.67 ± 4% 26.76 ± 8% +29.43% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 76.62 ± 12% 57.59 ± 7% -24.85% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 114.3 ± 5% 107.1 ± 5% ~ (p=0.065 n=6) RIBFlush/100000_routes,_5_peers-36 218.8 ± 10% 221.4 ± 11% ~ (p=0.818 n=6) geomean 10.84 12.37 +14.15% ```

It looks like the second one is more optimized on the paper, but the insertion rate is terrible. ``` goos: linux goarch: amd64 pkg: akvorado/outlet/routing/provider/bmp cpu: Intel(R) Xeon(R) W-2295 CPU @ 3.00GHz │ bench-step0.txt │ bench-step10.txt │ │ bytes/route │ bytes/route vs base │ RIBInsertion/1000_routes,_1_peers-36 345.5 ± 4% 462.2 ± 9% +33.76% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 360.0 ± 3% 440.8 ± 1% +22.43% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 323.5 ± 0% 436.0 ± 0% +34.78% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 322.8 ± 0% 437.4 ± 0% +35.51% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 317.9 ± 0% 428.8 ± 0% +34.87% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 269.8 ± 0% 427.5 ± 0% +58.45% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 265.0 ± 0% 420.5 ± 0% +58.68% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 251.0 ± 0% 398.8 ± 0% +58.88% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 207.1 ± 1% 346.8 ± 0% +67.46% (p=0.002 n=6) geomean 291.8 420.8 +44.21% │ bench-step0.txt │ bench-step10.txt │ │ sec/route │ sec/route vs base │ RIBInsertion/1000_routes,_1_peers-36 1.135µ ± 9% 10.740µ ± 6% +846.26% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 1.133µ ± 5% 10.124µ ± 6% +793.95% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 1.112µ ± 5% 11.789µ ± 6% +960.12% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 1.189µ ± 8% 18.016µ ± 13% +1415.82% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 1.242µ ± 12% 26.777µ ± 9% +2056.87% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 1.205µ ± 7% 50.710µ ± 6% +4110.05% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 1.086µ ± 8% 86.169µ ± 3% +7838.23% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 1.037µ ± 7% 164.565µ ± 4% +15776.94% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 1.154µ ± 6% 651.863µ ± 3% +56411.75% (p=0.002 n=6) geomean 1.142µ 40.45µ +3442.75% │ bench-step0.txt │ bench-step10.txt │ │ allocs/op │ allocs/op vs base │ RIBInsertion/1000_routes,_1_peers-36 2.069k ± 0% 58.643k ± 0% +2734.36% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 3.376k ± 0% 79.115k ± 0% +2243.47% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 6.900k ± 0% 155.886k ± 1% +2159.22% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 13.23k ± 0% 352.27k ± 1% +2562.90% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 27.52k ± 0% 923.95k ± 2% +3257.61% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 81.70k ± 0% 3572.06k ± 0% +4272.32% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 183.0k ± 0% 8565.2k ± 0% +4581.69% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 354.7k ± 0% 15583.8k ± 0% +4293.85% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 690.1k ± 0% 23956.4k ± 0% +3371.40% (p=0.002 n=6) RIBLookup/1000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_5_peers-36 0.000 ± 0% RIBFlush/1000_routes,_1_peers-36 2.055k ± 0% RIBFlush/1000_routes,_2_peers-36 3.363k ± 0% RIBFlush/1000_routes,_5_peers-36 6.834k ± 0% RIBFlush/10000_routes,_1_peers-36 13.01k ± 0% RIBFlush/10000_routes,_2_peers-36 27.13k ± 0% RIBFlush/10000_routes,_5_peers-36 81.30k ± 0% RIBFlush/100000_routes,_1_peers-36 181.4k ± 0% RIBFlush/100000_routes,_2_peers-36 352.1k ± 0% RIBFlush/100000_routes,_5_peers-36 685.2k ± 0% RTAHash-36 0.000 ± 0% RTAEqual-36 0.000 ± 0% geomean ² +379.85% ³ ² ¹ all samples are equal ² summaries must be >0 to compute geomean ³ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step10.txt │ │ sec/op │ sec/op vs base │ RIBLookup/1000_routes,_1_peers-36 30.39n ± 5% 29.89n ± 3% ~ (p=0.485 n=6) RIBLookup/1000_routes,_2_peers-36 30.30n ± 4% 30.72n ± 2% ~ (p=0.180 n=6) RIBLookup/1000_routes,_5_peers-36 32.59n ± 21% 32.66n ± 3% ~ (p=1.000 n=6) RIBLookup/10000_routes,_1_peers-36 33.58n ± 13% 35.03n ± 5% ~ (p=0.240 n=6) RIBLookup/10000_routes,_2_peers-36 37.09n ± 5% 37.12n ± 4% ~ (p=0.851 n=6) RIBLookup/10000_routes,_5_peers-36 44.42n ± 12% 42.56n ± 5% ~ (p=0.180 n=6) RIBLookup/100000_routes,_1_peers-36 88.08n ± 15% 97.26n ± 2% ~ (p=0.065 n=6) RIBLookup/100000_routes,_2_peers-36 96.64n ± 6% 111.25n ± 2% +15.12% (p=0.002 n=6) RIBLookup/100000_routes,_5_peers-36 102.5n ± 9% 121.2n ± 4% +18.29% (p=0.002 n=6) RIBLookup/1000000_routes,_1_peers-36 159.0n ± 8% 177.3n ± 33% +11.51% (p=0.002 n=6) RIBLookup/1000000_routes,_2_peers-36 157.9n ± 10% 174.1n ± 8% +10.22% (p=0.015 n=6) RIBLookup/1000000_routes,_5_peers-36 181.2n ± 17% RTAHash-36 19.30n ± 4% RTAEqual-36 8.700n ± 9% geomean 52.17n 63.46n +5.73% ¹ ¹ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ │ ms/op │ RIBFlush/1000_routes,_1_peers-36 578.6m ± 8% RIBFlush/1000_routes,_2_peers-36 922.9m ± 8% RIBFlush/1000_routes,_5_peers-36 1.825 ± 4% RIBFlush/10000_routes,_1_peers-36 5.619 ± 8% RIBFlush/10000_routes,_2_peers-36 9.527 ± 4% RIBFlush/10000_routes,_5_peers-36 20.67 ± 4% RIBFlush/100000_routes,_1_peers-36 76.62 ± 12% RIBFlush/100000_routes,_2_peers-36 114.3 ± 5% RIBFlush/100000_routes,_5_peers-36 218.8 ± 10% geomean 10.84 ```

Insertion is far too slow with pb.MapOf. Keep some of the improvements: - use RangeRelaxed() instead of Range() as we already have a writer lock - keep LoadAndDelete() - keep Compute() This is the final commit of this series. 42% increase in space, 1000% increase in insertion time, 12% lookup. The benchmarks do not test performance of multiple routines updating/looking up the RIB, but the idea is that it should behave better without locks, even if it is slower during updates. ``` goos: linux goarch: amd64 pkg: akvorado/outlet/routing/provider/bmp cpu: Intel(R) Xeon(R) W-2295 CPU @ 3.00GHz │ bench-step0.txt │ bench-step11.txt │ │ bytes/route │ bytes/route vs base │ RIBInsertion/1000_routes,_1_peers-36 345.5 ± 4% 453.9 ± 6% +31.37% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 360.0 ± 3% 434.4 ± 1% +20.68% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 323.5 ± 0% 442.4 ± 0% +36.77% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 322.8 ± 0% 440.6 ± 0% +36.51% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 317.9 ± 0% 432.5 ± 0% +36.05% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 269.8 ± 0% 418.1 ± 0% +54.99% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 265.0 ± 0% 411.3 ± 0% +55.21% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 251.0 ± 0% 389.5 ± 0% +55.18% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 207.1 ± 1% 338.5 ± 0% +63.45% (p=0.002 n=6) geomean 291.8 416.5 +42.72% │ bench-step0.txt │ bench-step11.txt │ │ sec/route │ sec/route vs base │ RIBInsertion/1000_routes,_1_peers-36 1.135µ ± 9% 9.961µ ± 6% +777.62% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 1.133µ ± 5% 8.840µ ± 6% +680.57% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 1.112µ ± 5% 9.631µ ± 11% +766.05% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 1.189µ ± 8% 11.728µ ± 14% +886.79% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 1.242µ ± 12% 13.114µ ± 11% +956.26% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 1.205µ ± 7% 15.885µ ± 7% +1218.80% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 1.086µ ± 8% 14.773µ ± 6% +1260.89% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 1.037µ ± 7% 14.493µ ± 6% +1298.22% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 1.154µ ± 6% 11.614µ ± 6% +906.81% (p=0.002 n=6) geomean 1.142µ 12.00µ +950.83% │ bench-step0.txt │ bench-step11.txt │ │ allocs/op │ allocs/op vs base │ RIBInsertion/1000_routes,_1_peers-36 2.069k ± 0% 58.389k ± 0% +2722.06% (p=0.002 n=6) RIBInsertion/1000_routes,_2_peers-36 3.376k ± 0% 78.648k ± 0% +2229.62% (p=0.002 n=6) RIBInsertion/1000_routes,_5_peers-36 6.900k ± 0% 155.323k ± 1% +2151.06% (p=0.002 n=6) RIBInsertion/10000_routes,_1_peers-36 13.23k ± 0% 349.77k ± 1% +2543.95% (p=0.002 n=6) RIBInsertion/10000_routes,_2_peers-36 27.52k ± 0% 923.58k ± 1% +3256.27% (p=0.002 n=6) RIBInsertion/10000_routes,_5_peers-36 81.70k ± 0% 3554.40k ± 1% +4250.71% (p=0.002 n=6) RIBInsertion/100000_routes,_1_peers-36 183.0k ± 0% 8533.5k ± 0% +4564.36% (p=0.002 n=6) RIBInsertion/100000_routes,_2_peers-36 354.7k ± 0% 15520.3k ± 0% +4275.96% (p=0.002 n=6) RIBInsertion/100000_routes,_5_peers-36 690.1k ± 0% 23823.9k ± 0% +3352.20% (p=0.002 n=6) RIBLookup/1000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/10000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/100000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_1_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_2_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBLookup/1000000_routes,_5_peers-36 0.000 ± 0% 0.000 ± 0% ~ (p=1.000 n=6) ¹ RIBFlush/1000_routes,_1_peers-36 2055.00 ± 0% 41.00 ± 0% -98.00% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 3.363k ± 0% 2.047k ± 0% -39.13% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 6.834k ± 0% 5.651k ± 0% -17.31% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 13011.0 ± 0% 118.0 ± 0% -99.09% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 27.13k ± 0% 12.98k ± 0% -52.17% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 81.30k ± 0% 62.19k ± 0% -23.50% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 181413.0 ± 0% 587.0 ± 0% -99.68% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 352.1k ± 0% 182.1k ± 0% -48.28% (p=0.002 n=6) RIBFlush/100000_routes,_5_peers-36 685.2k ± 0% 591.8k ± 0% -13.62% (p=0.002 n=6) RTAHash-36 0.000 ± 0% RTAEqual-36 0.000 ± 0% geomean ² +62.03% ³ ² ¹ all samples are equal ² summaries must be >0 to compute geomean ³ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step11.txt │ │ sec/op │ sec/op vs base │ RIBLookup/1000_routes,_1_peers-36 30.39n ± 5% 31.13n ± 6% ~ (p=0.818 n=6) RIBLookup/1000_routes,_2_peers-36 30.30n ± 4% 28.74n ± 7% -5.16% (p=0.026 n=6) RIBLookup/1000_routes,_5_peers-36 32.59n ± 21% 31.65n ± 4% ~ (p=0.093 n=6) RIBLookup/10000_routes,_1_peers-36 33.58n ± 13% 35.00n ± 2% ~ (p=0.240 n=6) RIBLookup/10000_routes,_2_peers-36 37.09n ± 5% 38.62n ± 3% +4.12% (p=0.041 n=6) RIBLookup/10000_routes,_5_peers-36 44.42n ± 12% 43.37n ± 6% ~ (p=0.818 n=6) RIBLookup/100000_routes,_1_peers-36 88.08n ± 15% 98.21n ± 3% ~ (p=0.065 n=6) RIBLookup/100000_routes,_2_peers-36 96.64n ± 6% 111.75n ± 3% +15.64% (p=0.002 n=6) RIBLookup/100000_routes,_5_peers-36 102.5n ± 9% 122.6n ± 4% +19.56% (p=0.002 n=6) RIBLookup/1000000_routes,_1_peers-36 159.0n ± 8% 173.5n ± 20% ~ (p=0.132 n=6) RIBLookup/1000000_routes,_2_peers-36 157.9n ± 10% 234.5n ± 24% +48.46% (p=0.002 n=6) RIBLookup/1000000_routes,_5_peers-36 181.2n ± 17% 294.4n ± 34% +62.46% (p=0.009 n=6) RTAHash-36 19.30n ± 4% RTAEqual-36 8.700n ± 9% geomean 52.17n 73.95n +12.38% ¹ ¹ benchmark set differs from baseline; geomeans may not be comparable │ bench-step0.txt │ bench-step11.txt │ │ ms/op │ ms/op vs base │ RIBFlush/1000_routes,_1_peers-36 578.6m ± 8% 728.7m ± 2% +25.94% (p=0.002 n=6) RIBFlush/1000_routes,_2_peers-36 922.9m ± 8% 1331.0m ± 3% +44.22% (p=0.002 n=6) RIBFlush/1000_routes,_5_peers-36 1.825 ± 4% 2.481 ± 3% +35.88% (p=0.002 n=6) RIBFlush/10000_routes,_1_peers-36 5.619 ± 8% 6.906 ± 3% +22.92% (p=0.002 n=6) RIBFlush/10000_routes,_2_peers-36 9.527 ± 4% 11.630 ± 8% +22.07% (p=0.002 n=6) RIBFlush/10000_routes,_5_peers-36 20.67 ± 4% 28.56 ± 5% +38.16% (p=0.002 n=6) RIBFlush/100000_routes,_1_peers-36 76.62 ± 12% 66.82 ± 9% -12.79% (p=0.002 n=6) RIBFlush/100000_routes,_2_peers-36 114.3 ± 5% 125.2 ± 9% +9.49% (p=0.009 n=6) RIBFlush/100000_routes,_5_peers-36 218.8 ± 10% 256.9 ± 4% +17.44% (p=0.002 n=6) geomean 10.84 13.16 +21.42% ```

This commit will be reverted at some point.

vincentbernat · 2026-02-24T06:45:51Z

See tests in #2092. This does not help.

vincentbernat added the service::outlet Impact the outlet service label Feb 17, 2026

This was referenced Feb 17, 2026

outlet/routing: buffer BMP messages to avoid being flagged as "stuck" #2245

Merged

Reduce lock contention when receiving a lot of BMP messages #2092

Closed

vincentbernat force-pushed the fix/bmp-lockless3 branch from 933c8b5 to c59d1c0 Compare February 21, 2026 07:07

vincentbernat added 11 commits February 22, 2026 08:15

outlet/routing: add benchmarks

3a736ab

This commit will be reverted at some point.

vincentbernat force-pushed the fix/bmp-lockless3 branch from c59d1c0 to 3a736ab Compare February 22, 2026 07:15

vincentbernat closed this Feb 24, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

outlet/routing: read from RIB without any locks#2244

outlet/routing: read from RIB without any locks#2244
vincentbernat wants to merge 11 commits into
mainfrom
fix/bmp-lockless3

vincentbernat commented Feb 17, 2026

Uh oh!

codecov Bot commented Feb 17, 2026 •

edited

Loading

Uh oh!

vincentbernat commented Feb 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

vincentbernat commented Feb 17, 2026

Uh oh!

codecov Bot commented Feb 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

vincentbernat commented Feb 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

codecov Bot commented Feb 17, 2026 •

edited

Loading