Standardize naming prefixes for kubernetes network objects by parulbajaj01 · Pull Request #3644 · GoogleCloudPlatform/cluster-toolkit

parulbajaj01 · 2025-02-06T13:07:56Z

Standardize naming convention for k8s network objects.

For A3H:

default
vpc1
...
vpc4

For A3M:

default
vpc1
...
vpc8

For A3U:

default
gvnic1
rdma0
...
rdma7

This aligns with the naming convention assumed in all Toolkit and gcloud docs, reducing friction experienced by users. This also avoids the need for users to rename networks in various kubernetes manifests, for e.g. the NCCL tests.

Submission Checklist

NOTE: Community submissions can take up to 2 weeks to be reviewed.

Please take the following actions before submitting this pull request.

Fork your PR branch from the Toolkit "develop" branch (not main)
Test all changes with pre-commit in a local branch #
Confirm that "make tests" passes all tests
Add or modify unit tests to cover code changes
Ensure that unit test coverage remains above 80%
Update all applicable documentation
Follow Cluster Toolkit Contribution guidelines #

annuay-google · 2025-02-14T12:26:49Z

The script path for all_gather NCCL test was broken, please fix it

annuay-google · 2025-02-14T12:28:26Z

Please update version for tcpxo NCCL test/installer to match what's in https://github.com/GoogleCloudPlatform/container-engine-accelerators repo

annuay-google

Requested changes to defaults, a missing bug-fix for running all_gather test and tcpxo version update

annuay-google

Requesting changes to defaults, a missing bug fix to run all_gather benchmark and tcpxo nccl version update

annuay-google

Changes LGTM. Let's merge after testing!

annuay-google · 2025-02-20T13:35:02Z

All the possible cluster provisioning paths have been tested with a NCCL test:

A3 high - toolkit instructions - verified
#
#                                                              out-of-place                       in-place          
#       size         count      type   redop    root     time   algbw   busbw #wrong     time   algbw   busbw #wrong
#        (B)    (elements)                               (us)  (GB/s)  (GB/s)            (us)  (GB/s)  (GB/s)       
     1048576         16384     float    none      -1   1116.2    0.94    0.88    N/A   1094.0    0.96    0.90    N/A
     2097152         32768     float    none      -1   1071.4    1.96    1.83    N/A   1072.8    1.95    1.83    N/A
     4194304         65536     float    none      -1   1123.8    3.73    3.50    N/A   1089.6    3.85    3.61    N/A
     8388608        131072     float    none      -1   1211.6    6.92    6.49    N/A   1209.0    6.94    6.50    N/A
    16777216        262144     float    none      -1   1263.4   13.28   12.45    N/A   1322.5   12.69   11.89    N/A
    33554432        524288     float    none      -1   1305.2   25.71   24.10    N/A   1304.5   25.72   24.11    N/A
    67108864       1048576     float    none      -1   1522.1   44.09   41.33    N/A   1522.6   44.07   41.32    N/A
   134217728       2097152     float    none      -1   2156.8   62.23   58.34    N/A   2217.9   60.52   56.73    N/A
   268435456       4194304     float    none      -1   3685.5   72.84   68.28    N/A   3588.1   74.81   70.14    N/A
   536870912       8388608     float    none      -1   7238.0   74.17   69.54    N/A   7018.7   76.49   71.71    N/A
# Out of bounds values : 0 OK
# Avg bus bandwidth    : 28.7753

A3 high - gcloud docs instructions - verified
#
#                                                              out-of-place                       in-place          
#       size         count      type   redop    root     time   algbw   busbw #wrong     time   algbw   busbw #wrong
#        (B)    (elements)                               (us)  (GB/s)  (GB/s)            (us)  (GB/s)  (GB/s)       
     1048576         16384     float    none      -1    935.0    1.12    1.05    N/A    954.1    1.10    1.03    N/A
     2097152         32768     float    none      -1    940.7    2.23    2.09    N/A    978.2    2.14    2.01    N/A
     4194304         65536     float    none      -1    997.9    4.20    3.94    N/A    995.3    4.21    3.95    N/A
     8388608        131072     float    none      -1   1048.8    8.00    7.50    N/A   1061.5    7.90    7.41    N/A
    16777216        262144     float    none      -1   1079.6   15.54   14.57    N/A   1148.7   14.60   13.69    N/A
    33554432        524288     float    none      -1   1172.8   28.61   26.82    N/A   1180.2   28.43   26.65    N/A
    67108864       1048576     float    none      -1   1355.5   49.51   46.41    N/A   1347.9   49.79   46.68    N/A
   134217728       2097152     float    none      -1   2044.4   65.65   61.55    N/A   2015.0   66.61   62.45    N/A
   268435456       4194304     float    none      -1   3626.5   74.02   69.39    N/A   3581.8   74.94   70.26    N/A
   536870912       8388608     float    none      -1   7058.4   76.06   71.31    N/A   7020.7   76.47   71.69    N/A
# Out of bounds values : 0 OK
# Avg bus bandwidth    : 30.5226

A3 mega - toolkit instructions - verified
#       size         count      type   redop    root     time   algbw   busbw #wrong     time   algbw   busbw #wrong
#        (B)    (elements)                               (us)  (GB/s)  (GB/s)            (us)  (GB/s)  (GB/s)       
           0             0     float    none      -1     0.23    0.00    0.00      0     0.17    0.00    0.00      0
           0             0     float    none      -1     0.16    0.00    0.00      0     0.16    0.00    0.00      0
           0             0     float    none      -1     0.16    0.00    0.00      0     0.16    0.00    0.00      0
           0             0     float    none      -1     0.17    0.00    0.00      0     0.16    0.00    0.00      0
           0             0     float    none      -1     0.16    0.00    0.00      0     0.16    0.00    0.00      0
         256             4     float    none      -1    238.9    0.00    0.00      0    241.6    0.00    0.00      0
         512             8     float    none      -1    238.3    0.00    0.00      0    238.7    0.00    0.00      0
        1024            16     float    none      -1    238.7    0.00    0.00      0    239.5    0.00    0.00      0
        2048            32     float    none      -1    237.0    0.01    0.01      0    238.9    0.01    0.01      0
        4096            64     float    none      -1    237.3    0.02    0.02      0    236.5    0.02    0.02      0
        8192           128     float    none      -1    237.8    0.03    0.03      0    237.4    0.03    0.03      0
       16384           256     float    none      -1    242.0    0.07    0.06      0    241.6    0.07    0.06      0
       32768           512     float    none      -1    240.2    0.14    0.13      0    241.4    0.14    0.13      0
       65536          1024     float    none      -1    263.6    0.25    0.23      0    264.1    0.25    0.23      0
      131072          2048     float    none      -1    265.5    0.49    0.46      0    281.4    0.47    0.44      0
      262144          4096     float    none      -1    285.1    0.92    0.86      0    292.8    0.90    0.84      0
      524288          8192     float    none      -1    315.7    1.66    1.56      0    317.3    1.65    1.55      0
     1048576         16384     float    none      -1    310.4    3.38    3.17      0    319.2    3.29    3.08      0
     2097152         32768     float    none      -1    308.9    6.79    6.36      0    316.7    6.62    6.21      0
     4194304         65536     float    none      -1    330.9   12.68   11.88      0    325.5   12.89   12.08      0
     8388608        131072     float    none      -1    356.3   23.55   22.07      0    353.1   23.76   22.27      0
    16777216        262144     float    none      -1    411.4   40.78   38.24      0    408.8   41.04   38.47      0
    33554432        524288     float    none      -1    479.9   69.92   65.55      0    479.8   69.93   65.56      0
    67108864       1048576     float    none      -1    732.7   91.59   85.86      0    730.7   91.84   86.10      0
   134217728       2097152     float    none      -1   1107.0  121.24  113.66      0   1097.1  122.33  114.69      0
   268435456       4194304     float    none      -1   1775.2  151.22  141.77      0   1777.8  151.00  141.56      0
   536870912       8388608     float    none      -1   2967.3  180.93  169.62      0   2971.7  180.66  169.37      0
  1073741824      16777216     float    none      -1   5552.6  193.38  181.29      0   5532.8  194.07  181.94      0
  2147483648      33554432     float    none      -1    10854  197.85  185.48      0    10835  198.19  185.80      0
  4294967296      67108864     float    none      -1    21502  199.75  187.26      0    21489  199.87  187.38      0
  8589934592     134217728     float    none      -1    42809  200.66  188.12      0    42774  200.82  188.27      0
# Out of bounds values : 0 OK
# Avg bus bandwidth    : 45.3197 
#

A3 mega - gcloud docs instructions - WIP
#                                                              out-of-place                       in-place          
#       size         count      type   redop    root     time   algbw   busbw #wrong     time   algbw   busbw #wrong
#        (B)    (elements)                               (us)  (GB/s)  (GB/s)            (us)  (GB/s)  (GB/s)       
           0             0     float    none      -1     0.24    0.00    0.00      0     0.16    0.00    0.00      0
           0             0     float    none      -1     0.17    0.00    0.00      0     0.15    0.00    0.00      0
           0             0     float    none      -1     0.17    0.00    0.00      0     0.15    0.00    0.00      0
           0             0     float    none      -1     0.15    0.00    0.00      0     0.15    0.00    0.00      0
           0             0     float    none      -1     0.16    0.00    0.00      0     0.17    0.00    0.00      0
         256             4     float    none      -1    234.3    0.00    0.00      0    236.8    0.00    0.00      0
         512             8     float    none      -1    233.9    0.00    0.00      0    231.2    0.00    0.00      0
        1024            16     float    none      -1    232.8    0.00    0.00      0    231.8    0.00    0.00      0
        2048            32     float    none      -1    231.8    0.01    0.01      0    233.9    0.01    0.01      0
        4096            64     float    none      -1    232.6    0.02    0.02      0    232.7    0.02    0.02      0
        8192           128     float    none      -1    233.5    0.04    0.03      0    234.4    0.03    0.03      0
       16384           256     float    none      -1    236.1    0.07    0.07      0    236.4    0.07    0.06      0
       32768           512     float    none      -1    235.3    0.14    0.13      0    236.4    0.14    0.13      0
       65536          1024     float    none      -1    261.4    0.25    0.24      0    261.9    0.25    0.23      0
      131072          2048     float    none      -1    263.6    0.50    0.47      0    275.5    0.48    0.45      0
      262144          4096     float    none      -1    274.2    0.96    0.90      0    274.7    0.95    0.89      0
      524288          8192     float    none      -1    317.5    1.65    1.55      0    310.5    1.69    1.58      0
     1048576         16384     float    none      -1    316.2    3.32    3.11      0    318.0    3.30    3.09      0
     2097152         32768     float    none      -1    320.2    6.55    6.14      0    311.7    6.73    6.31      0
     4194304         65536     float    none      -1    326.3   12.85   12.05      0    333.6   12.57   11.79      0
     8388608        131072     float    none      -1    350.2   23.95   22.46      0    347.3   24.16   22.65      0
    16777216        262144     float    none      -1    408.1   41.11   38.54      0    400.9   41.85   39.24      0
    33554432        524288     float    none      -1    475.6   70.55   66.14      0    470.7   71.28   66.82      0
    67108864       1048576     float    none      -1    722.6   92.87   87.07      0    737.4   91.01   85.32      0
   134217728       2097152     float    none      -1   1111.1  120.80  113.25      0   1107.7  121.17  113.60      0
   268435456       4194304     float    none      -1   1730.4  155.13  145.43      0   1718.6  156.20  146.44      0
   536870912       8388608     float    none      -1   2869.5  187.09  175.40      0   2876.6  186.63  174.97      0
  1073741824      16777216     float    none      -1   5512.5  194.78  182.61      0   5536.8  193.93  181.81      0
  2147483648      33554432     float    none      -1    10843  198.04  185.67      0    10822  198.45  186.04      0
  4294967296      67108864     float    none      -1    21450  200.23  187.72      0    21594  198.90  186.47      0
  8589934592     134217728     float    none      -1    42672  201.30  188.72      0    42654  201.38  188.80      0
# Out of bounds values : 0 OK
# Avg bus bandwidth    : 45.7171

A3 ultra - gcloud toolkit instructions - verified
#                                                              out-of-place                       in-place          
#       size         count      type   redop    root     time   algbw   busbw #wrong     time   algbw   busbw #wrong
#        (B)    (elements)                               (us)  (GB/s)  (GB/s)            (us)  (GB/s)  (GB/s)       
        1024            16     float    none      -1    44.22    0.02    0.02      0    43.28    0.02    0.02      0
        2048            32     float    none      -1    43.02    0.05    0.04      0    43.17    0.05    0.04      0
        4096            64     float    none      -1    43.09    0.10    0.09      0    43.02    0.10    0.09      0
        8192           128     float    none      -1    43.38    0.19    0.18      0    43.48    0.19    0.18      0
       16384           256     float    none      -1    44.04    0.37    0.35      0    44.16    0.37    0.35      0
       32768           512     float    none      -1    45.15    0.73    0.68      0    45.02    0.73    0.68      0
       65536          1024     float    none      -1    47.33    1.38    1.30      0    47.14    1.39    1.30      0
      131072          2048     float    none      -1    47.30    2.77    2.60      0    49.76    2.63    2.47      0
      262144          4096     float    none      -1    51.89    5.05    4.74      0    50.20    5.22    4.90      0
      524288          8192     float    none      -1    54.76    9.57    8.98      0    53.83    9.74    9.13      0
     1048576         16384     float    none      -1    72.77   14.41   13.51      0    72.46   14.47   13.57      0
     2097152         32768     float    none      -1    76.52   27.40   25.69      0    74.89   28.00   26.25      0
     4194304         65536     float    none      -1    82.02   51.14   47.94      0    80.70   51.98   48.73      0
     8388608        131072     float    none      -1    91.48   91.70   85.97      0    93.68   89.54   83.95      0
    16777216        262144     float    none      -1    121.2  138.39  129.74      0    121.7  137.87  129.26      0
    33554432        524288     float    none      -1    182.4  183.98  172.48      0    178.9  187.61  175.89      0
    67108864       1048576     float    none      -1    280.4  239.32  224.36      0    279.6  239.98  224.98      0
   134217728       2097152     float    none      -1    497.3  269.90  253.03      0    490.5  273.61  256.51      0
   268435456       4194304     float    none      -1    864.0  310.70  291.28      0    853.7  314.45  294.80      0
   536870912       8388608     float    none      -1   1556.4  344.95  323.39      0   1552.0  345.92  324.30      0
  1073741824      16777216     float    none      -1   3083.6  348.21  326.45      0   3041.0  353.09  331.02      0
  2147483648      33554432     float    none      -1   6135.2  350.03  328.15      0   6035.9  355.79  333.55      0
  4294967296      67108864     float    none      -1    12239  350.92  328.99      0    11961  359.09  336.65      0
  8589934592     134217728     float    none      -1    24403  352.01  330.01      0    23913  359.21  336.76      0
# Out of bounds values : 0 OK
# Avg bus bandwidth    : 121.569 
#

A3 ultra - gcloud custom cluster instructions (2 node) - verified
#                                                              out-of-place                       in-place          
#       size         count      type   redop    root     time   algbw   busbw #wrong     time   algbw   busbw #wrong
#        (B)    (elements)                               (us)  (GB/s)  (GB/s)            (us)  (GB/s)  (GB/s)       
        1024            16     float    none      -1    45.61    0.02    0.02      0    44.73    0.02    0.02      0
        2048            32     float    none      -1    44.93    0.05    0.04      0    44.73    0.05    0.04      0
        4096            64     float    none      -1    45.31    0.09    0.08      0    45.28    0.09    0.08      0
        8192           128     float    none      -1    45.25    0.18    0.17      0    45.58    0.18    0.17      0
       16384           256     float    none      -1    46.55    0.35    0.33      0    46.37    0.35    0.33      0
       32768           512     float    none      -1    48.21    0.68    0.64      0    47.47    0.69    0.65      0
       65536          1024     float    none      -1    49.49    1.32    1.24      0    50.21    1.31    1.22      0
      131072          2048     float    none      -1    49.04    2.67    2.51      0    52.31    2.51    2.35      0
      262144          4096     float    none      -1    50.85    5.16    4.83      0    51.11    5.13    4.81      0
      524288          8192     float    none      -1    55.68    9.42    8.83      0    56.90    9.21    8.64      0
     1048576         16384     float    none      -1    75.04   13.97   13.10      0    73.83   14.20   13.32      0
     2097152         32768     float    none      -1    76.15   27.54   25.82      0    76.77   27.32   25.61      0
     4194304         65536     float    none      -1    78.74   53.27   49.94      0    78.70   53.30   49.97      0
     8388608        131072     float    none      -1    94.29   88.96   83.40      0    91.44   91.73   86.00      0
    16777216        262144     float    none      -1    122.3  137.20  128.63      0    124.0  135.28  126.82      0
    33554432        524288     float    none      -1    181.4  184.93  173.37      0    181.4  185.01  173.45      0
    67108864       1048576     float    none      -1    280.6  239.20  224.25      0    276.1  243.03  227.84      0
   134217728       2097152     float    none      -1    498.5  269.22  252.40      0    494.8  271.28  254.33      0
   268435456       4194304     float    none      -1    861.7  311.50  292.03      0    857.8  312.94  293.38      0
   536870912       8388608     float    none      -1   1558.0  344.60  323.06      0   1556.1  345.02  323.45      0
  1073741824      16777216     float    none      -1   3082.3  348.36  326.59      0   3046.2  352.48  330.45      0
  2147483648      33554432     float    none      -1   6142.6  349.61  327.76      0   6037.7  355.68  333.45      0
  4294967296      67108864     float    none      -1    12235  351.05  329.11      0    12002  357.87  335.50      0
  8589934592     134217728     float    none      -1    24433  351.57  329.59      0    23902  359.38  336.92      0
# Out of bounds values : 0 OK
# Avg bus bandwidth    : 121.386 
#

A3 ultra - gcloud custom cluster instructions (Jobset) - verified
Broken, but expected - these instructions use the TAS plugin

annuay-google

LGTM

…for tcpxo

parulbajaj01 added the release-improvements Added to release notes under the "Improvements" heading. label Feb 6, 2025

parulbajaj01 requested a review from annuay-google February 6, 2025 13:07

parulbajaj01 marked this pull request as draft February 6, 2025 13:08

parulbajaj01 force-pushed the gke-network-config branch from ae76095 to efd436d Compare February 13, 2025 13:57