Here are some benchmarks for PPA Lustre filesystem that was upgraded in March 2014.

System specs: 

  • Lustre server version 2.5.1 on RHEL6.5
  • 1 MDS and 8 OSS servers using Dell R610 systems
  • Four LSI Engenio (Dell MD3260) arrays with dual-redundant controllers
  • 16 OSTs using RAID6 LUNs with 2TB NL SAS drives
  • ~344TB of usable capacity
  • 10Gb ethernet links to SLAC network
  • QDR Infiniband links to "bullet" PPA compute cluster

IOR is a parallel I/O benchmarking utility that runs MPI-coordinated file transfers. The test clients are "bullet" cluster nodes with Lustre mounted over QDR Infiniband. Maximum throughput is > 6GB/sec using multiple nodes or ~3GB/sec with a single node. 

1 process on 1 node

Max Write: 419.58 MiB/sec (439.96 MB/sec)
Max Read:  399.31 MiB/sec (418.71 MB/sec)

Command line used: src/C/IOR -F -a POSIX -r -w -b 60g -t 1m -o /lustre/ki/pfs/yemi/IORFile -O lustreStripeCount=0 -v -i 2

       api                = POSIX
       test filename      = /lustre/ki/pfs/yemi/IORFile
       access             = file-per-process
       pattern            = segmented (1 segment)
       ordering in a file = sequential offsets
       ordering inter file= no tasks offsets
       clients            = 1 (1 per node)
       repetitions        = 2
       xfersize           = 1 MiB
       blocksize          = 60 GiB
       aggregate filesize = 60 GiB
       Lustre stripe size = Use default
             stripe count = Use default

Operation  Max (MiB)  Min (MiB)  Mean (MiB)   Std Dev  Max (OPs)  Min (OPs)  Mean (OPs)   Std Dev  Mean (s)  Op grep #Tasks tPN reps  fPP reord reordoff reordrand seed segcnt blksiz xsize aggsize

---------  ---------  ---------  ----------   -------  ---------  ---------  ----------   -------  --------
write         419.58     412.07      415.82      3.75     419.58     412.07      415.82      3.75 147.76775   1 1 2 1 0 1 0 0 1 64424509440 1048576 64424509440 -1 POSIX EXCEL
read          399.31     391.06      395.18      4.13     399.31     391.06      395.18      4.13 155.48960   1 1 2 1 0 1 0 0 1 64424509440 1048576 64424509440 -1 POSIX EXCEL

3 processes on 1 node

Max Write: 1133.49 MiB/sec (1188.55 MB/sec)
Max Read:  1055.04 MiB/sec (1106.29 MB/sec)

Command line used: src/C/IOR -F -a POSIX -r -w -b 20g -t 1m -o /lustre/ki/pfs/yemi/IORFile -O lustreStripeCount=0 -v -i 2

       api                = POSIX
       test filename      = /lustre/ki/pfs/yemi/IORFile
       access             = file-per-process
       pattern            = segmented (1 segment)
       ordering in a file = sequential offsets
       ordering inter file= no tasks offsets
       clients            = 3 (3 per node)
       repetitions        = 2
       xfersize           = 1 MiB
       blocksize          = 20 GiB
       aggregate filesize = 60 GiB
       Lustre stripe size = Use default
             stripe count = Use default

Operation  Max (MiB)  Min (MiB)  Mean (MiB)   Std Dev  Max (OPs)  Min (OPs)  Mean (OPs)   Std Dev  Mean (s)  Op grep #Tasks tPN reps  fPP reord reordoff reordrand seed segcnt blksiz xsize aggsize

---------  ---------  ---------  ----------   -------  ---------  ---------  ----------   -------  --------
write        1133.49     975.77     1054.63     78.86    1133.49     975.77     1054.63     78.86  58.58496   3 3 2 1 0 1 0 0 1 21474836480 1048576 64424509440 -1 POSIX EXCEL
read         1055.04     878.41      966.73     88.32    1055.04     878.41      966.73     88.32  64.08962   3 3 2 1 0 1 0 0 1 21474836480 1048576 64424509440 -1 POSIX EXCEL

8 processes on 1 node

Max Write: 2373.15 MiB/sec (2488.43 MB/sec)
Max Read:  2470.09 MiB/sec (2590.08 MB/sec)

Command line used: src/C/IOR -F -a POSIX -r -w -b 20g -t 1m -o /lustre/ki/pfs/yemi/IORFile -O lustreStripeCount=0 -v -i 2

 api                = POSIX
 test filename      = /lustre/ki/pfs/yemi/IORFile
 access             = file-per-process
 pattern            = segmented (1 segment)
 ordering in a file = sequential offsets
 ordering inter file= no tasks offsets
 clients            = 8 (8 per node)
 repetitions        = 2
 xfersize           = 1 MiB
 blocksize          = 20 GiB
 aggregate filesize = 160 GiB
 Lustre stripe size = Use default
       stripe count = Use default

Operation  Max (MiB)  Min (MiB)  Mean (MiB)   Std Dev  Max (OPs)  Min (OPs)  Mean (OPs)   Std Dev  Mean (s)  Op grep #Tasks tPN reps  fPP reord reordoff reordrand seed segcnt blksiz xsize aggsize

---------  ---------  ---------  ----------   -------  ---------  ---------  ----------   -------  --------
write        2373.15    2116.37     2244.76    128.39    2373.15    2116.37     2244.76    128.39  73.22739   8 8 2 1 0 1 0 0 1 21474836480 1048576 171798691840 -1 POSIX EXCEL
read         2470.09    2289.77     2379.93     90.16    2470.09    2289.77     2379.93     90.16  68.94123   8 8 2 1 0 1 0 0 1 21474836480 1048576 171798691840 -1 POSIX EXCEL

16 processes on 1 node

Max Write: 2959.63 MiB/sec (3103.40 MB/sec)
Max Read:  2456.74 MiB/sec (2576.08 MB/sec)

Command line used: src/C/IOR -F -a POSIX -r -w -b 20g -t 1m -o /lustre/ki/pfs/yemi/IORFile -O lustreStripeCount=0 -v -i 2

 api                = POSIX
 test filename      = /lustre/ki/pfs/yemi/IORFile
 access             = file-per-process
 pattern            = segmented (1 segment)
 ordering in a file = sequential offsets
 ordering inter file= no tasks offsets
 clients            = 16 (16 per node)
 repetitions        = 2
 xfersize           = 1 MiB
 blocksize          = 20 GiB
 aggregate filesize = 320 GiB
 Lustre stripe size = Use default
       stripe count = Use default

Operation  Max (MiB)  Min (MiB)  Mean (MiB)   Std Dev  Max (OPs)  Min (OPs)  Mean (OPs)   Std Dev  Mean (s)  Op grep #Tasks tPN reps  fPP reord reordoff reordrand seed segcnt blksiz xsize aggsize

---------  ---------  ---------  ----------   -------  ---------  ---------  ----------   -------  --------
write        2959.63    2767.42     2863.53     96.10    2959.63    2767.42     2863.53     96.10 114.56132   16 16 2 1 0 1 0 0 1 21474836480 1048576 343597383680 -1 POSIX EXCEL
read         2456.74    2430.77     2443.76     12.98    2456.74    2430.77     2443.76     12.98 134.09235   16 16 2 1 0 1 0 0 1 21474836480 1048576 343597383680 -1 POSIX EXCEL

8 processes across 4 nodes

Max Write: 3101.05 MiB/sec (3251.68 MB/sec)

 Max Read:  2917.03 MiB/sec (3058.73 MB/sec)


Command line used: src/C/IOR -F -a POSIX -r -w -b 30g -t 1m -o /lustre/ki/pfs/yemi/IORFile -O lustreStripeCount=0 -v -i 2

       api                = POSIX
       test filename      = /lustre/ki/pfs/yemi/IORFile
       access             = file-per-process
       pattern            = segmented (1 segment)
       ordering in a file = sequential offsets
       ordering inter file= no tasks offsets
       clients            = 8 (2 per node)
       repetitions        = 2
       xfersize           = 1 MiB
       blocksize          = 30 GiB
       aggregate filesize = 240 GiB
       Lustre stripe size = Use default
             stripe count = Use default

Operation  Max (MiB)  Min (MiB)  Mean (MiB)   Std Dev  Max (OPs)  Min (OPs)  Mean (OPs)   Std Dev  Mean (s)  Op grep #Tasks tPN reps  fPP reord reordoff reordrand seed segcnt blksiz xsize aggsize

---------  ---------  ---------  ----------   -------  ---------  ---------  ----------   -------  --------
write        3101.05    2873.49     2987.27    113.78    3101.05    2873.49     2987.27    113.78  82.38866   8 2 2 1 0 1 0 0 1 32212254720 1048576 257698037760 -1 POSIX EXCEL
read         2917.03    2749.41     2833.22     83.81    2917.03    2749.41     2833.22     83.81  86.81822   8 2 2 1 0 1 0 0 1 32212254720 1048576 257698037760 -1 POSIX EXCEL


32 processes across 8 nodes

Max Write: 5004.27 MiB/sec (5247.35 MB/sec)
Max Read:  6233.60 MiB/sec (6536.40 MB/sec)

Command line used: src/C/IOR -F -a POSIX -r -w -b 20g -t 1m -o /lustre/ki/pfs/yemi/IORFile -O lustreStripeCount=0 -v -i 2

       api                = POSIX
       test filename      = /lustre/ki/pfs/yemi/IORFile
       access             = file-per-process
       pattern            = segmented (1 segment)
       ordering in a file = sequential offsets
       ordering inter file= no tasks offsets
       clients            = 32 (4 per node)
       repetitions        = 2
       xfersize           = 1 MiB
       blocksize          = 20 GiB
       aggregate filesize = 640 GiB
       Lustre stripe size = Use default
             stripe count = Use default

Operation  Max (MiB)  Min (MiB)  Mean (MiB)   Std Dev  Max (OPs)  Min (OPs)  Mean (OPs)   Std Dev  Mean (s)  Op grep #Tasks tPN reps  fPP reord reordoff reordrand seed segcnt blksiz xsize aggsize

---------  ---------  ---------  ----------   -------  ---------  ---------  ----------   -------  --------
write        5004.27    4948.77     4976.52     27.75    5004.27    4948.77     4976.52     27.75 131.69457   32 4 2 1 0 1 0 0 1 21474836480 1048576 687194767360 -1 POSIX EXCEL
read         6233.60    6087.79     6160.70     72.91    6233.60    6087.79     6160.70     72.91 106.39250   32 4 2 1 0 1 0 0 1 21474836480 1048576 687194767360 -1 POSIX EXCEL

  • No labels