TWiki> UCSDTier2 Web>CMSLoadTestLHCONE>CMSLoadTestLHCONEResults (2011/06/04, JamesLetts)EditAttach

Test Results

Sites marked by an asterisk are connected to the LHCONE network when the test took place.

Our script calculates the maximum rate by taking a moving hourly average over 12 x 5 minute bins. Note that 1280 MiB/s corresponds to 10Gbps and 128 MiB/s to 1Gbps. The binning may make it appear that the the maximum transfer rate is higher than the line speed in some cases, due to rounding.

SOURCE SITE DESTINATION SITE Files Injected Start Time Max. Rate in 1h (MiB/s) Latency (h) Quality (%) Comment
T2_US_Wisconsin T2_US_MIT 1000 1305328040 859.1 4.9 97.5%  
T2_US_Wisconsin T2_US_MIT 2000 1305344878 1455.1 7.2 96.8%  
T2_US_Wisconsin T2_US_MIT 2000 1305396650 1379.4 12.8 88.8% Did not suspend other debug transfers.
T2_US_Wisconsin T2_US_MIT 2000 1305440850 1455.1 6.5 96.3% Did not suspend other debug transfers.
T2_US_Wisconsin T2_US_MIT 2000 1305614014 1050.4 217.4 62.3% Did not suspend other debug transfers.
T2_UK_London_IC T2_US_Purdue 1000 1305655527 526.9 287.9 94.9% Did not suspend significant production transfers. Purdue site went down.
T2_FR_GRIF_LLR T2_DE_RWTH 1000 1305921458 654.4 3.0 96.7%  
T2_DE_RWTH T2_FR_GRIF_LLR 1000 1305937471 640.8 6.6 100.0%  
T2_DE_RWTH T2_US_MIT 1000 1306351647 601.6 90.5 88.3% After 989 files, all transfers from RWTH fail with [GENERAL_FAILURE] AsyncWait and it appears the proxy at MIT has expired.
T2_IT_Pisa T2_IT_Legnaro 500 1306447372 303.3   95.5% Halted after 80% completed
T2_FR_GRIF_LLR T2_UK_London_IC 1000 1306758701 319.9 7.0 84.0%  
T2_ES_IFCA T2_IT_Pisa 1000 1306762972 116.5 19.5 97.6%  
T2_US_MIT T2_IT_Legnaro 1000 1306840432 136.7 9.4 97.9%  
T2_IT_Pisa T2_US_Wisconsin 1000 1306843798 161.2 28.8 60.4%  
T2_IT_Legnaro T2_US_MIT 1000 1306876968 179.4 5.0 100.0%  
T2_US_MIT T2_DE_RWTH 1000 1306898705 438.7 3.4 92.4%  
T2_UK_London_IC T2_FR_GRIF_LLR 1000 1306919225 93.9 18.6 99.0%  
T2_IT_Pisa T2_ES_IFCA 1000 1306923212 80.6 20.1 98.6%  
T2_DE_RWTH T2_US_MIT 1000 1306915439 263.3 5.6 98.7%  
T2_US_MIT T2_DE_DESY 1000 1306958130 89.5 10.5 99.9%  
T2_DE_DESY T2_US_MIT 1000 1306996242 277.8 6.7 100.0%  
T2_US_Wisconsin T2_DE_RWTH 1000 1306999306 472.9 6.5 93.9%  
T2_US_Purdue T2_IT_Legnaro 1000 1306998314 78.2 21.7 76.4%  
T2_RU_RRC_KI T2_IT_Pisa 500 1306996437 50.5 17.9 96.7%  
T2_DE_RWTH T2_US_Wisconsin 1000 1307032216 205.0 4.3 94.5%  
T2_US_Purdue T2_US_MIT 2000 1307057388 330.7 6.5 98.2%  
T2_US_MIT T2_IT_Pisa 1000 1307115239       No files transferred in first 12h
T2_DE_RWTH T2_UK_London_IC 1000 1307114126 248.6 6.8 83.8%  
T2_ES_IFCA T2_DE_DESY 1000 1307113378 85.7   100.0% 83% complete
T2_US_Wisconsin T2_IT_Legnaro 1000 1307125203 138.8   98.7% 97% complete

Observations:

There are three fundamental limitations to PhEDEx transfer rates, any of which may be the limiting factor in the maximum hourly transfer rate:

  1. FTS channel configuration (number of simultaneous file transfers, streams per transfer, queue length)
  2. PhEDEx download agent configuration (number of simultaneous file transfers)
  3. Network bandwidth (1-20 Gbps)

When the results above are ordered by destination site, then the limitation may become evident.

For example, MIT can pull data at wire speed from Wisconsin but not from Europe, where network speed may be the limiting factor.

Action Item List:

  • Developed a PhEDEx data service-based tool to get the metric numbers, especially the transfer latency. Script can be found here.

-- JamesLetts - 2011/05/13

Edit | Attach | Print version | History: r35 < r34 < r33 < r32 < r31 | Backlinks | Raw View | Raw edit | More topic actions
Topic revision: r35 - 2011/06/04 - 03:05:19 - JamesLetts
 
This site is powered by the TWiki collaboration platformCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback