Monitoring for T0&T1 Data Transfer tests

March 2017

Transfer Quality

Transfer Rates

T1_US_FNAL_MSS: 1054 MB/s T1_US_FNAL_Disk: 2400 MB/s
Ticket: https://ggus.eu/?mode=ticket_info&ticket_id=127830 Tape monitoring: https://ggus.eu/index.php?mode=download&attid=ATT107455

T0 Disk -> FNAL Disk (T0_CH_CERN_Disk to T1_US_FNAL_Disk graphic)

For the last test with and injection rate of 2400 MB/s (Apr 11). A peak rate of 2580 MB/s was reached and an average rate of ~2300MB/s The targets were reach in each iteration of the test.

T0 Disk -> FNAL MSS (T0_CH_CERN_Disk to T1_US_FNAL_Buffer graphic)

Peak rate of 2000 MB/s Average rate 1100 MB/s

FNAL MSS -> FNAL DISK (T1_US_FNAL_Buffer to T1_US_FNAL_Disk graphic)

The staging test was performed twice

Based on Phedex plots, the rates from Buffer to Disk were:

  • 1st test -> 1200 MB/s(peak rate) 325 MB/s (average).
  • 2nd test -> 821 MB/s (peak rate) 215 MB/s (average).

According to internal monitoring for the staging tests The peak reached was 1.8GB/s The individual transfer rate from tape to dCache pool is about 160MB/s. One newer node (in test configuration) reads at 256MB/s which is the maximal tape drive speed.

Information available via PhEDEx monitoring is consistent with internal monitoring behaviour.

T1_IT_CNAF_MSS: 486 MB/s T1_IT_CNAF_Disk: 1000 MB/s
Ticket: https://ggus.eu/?mode=ticket_info&ticket_id=127239

- T0 Disk -> Disk
Got a pick rate of ~800 MB/s, and an avg rate of 400MB/s
- Buffer -> Disk
Got a pick rate of ~600MB/s, and an avg rate of ~450MB/s


We conclude that the tests were successful, transfers from T0 Disk -> Disk endpoint gave and clear idea of how fast the data can be transferred to the site, since there were no target rates, understanding the link performance allows to be better plan the movement of data from T0 to T1 sites.
Regarding Buffer -> Disk transfers, tests were successful as well, since we also got and clear idea of the staging capabilities of the site. The plots from site admins and Phedex were consistent and this might have to do with the fact that at the moment only CMS transfers were taking place (needs to be confirmed).
Staging and data transfers can be optimized by serializing the files and enabling big tapes and better tuning Phedex agents can improve the staging from tape. Definitely starting to transfer the information in advance is a key item to take into account, as well as understanding the data distribution in the sites, data priority and deadlines.

T1_UK_RAL_MSS: 175 MB/s T1_UK_RAL_Disk: 1000 MB/s
Ticket: https://ggus.eu/?mode=ticket_info&ticket_id=127240 Tape monitoring:

- T0 Disk -> Disk Got a pick rate of ~600 MB/s, and an avg rate of 400MB/s

- Buffer -> Disk Got an avg rate of ~600MB/s

We conclude that the tests were successful, transfers from T0 Disk -> Disk endpoint gave and clear idea of how fast the data can be transferred to the site, since there were no target rates, understanding the link performance allows to be better plan the movement of data from T0 to T1 sites.
Regarding Buffer -> Disk transfers, tests were successful as well, since we also got and clear idea of the staging capabilities of the site. The plots from site admins and Phedex were consistent and this might have to do with the fact that at the moment only CMS transfers were taking place (needs to be confirmed).
Staging and data transfers can be optimized by serializing the files and enabling big tapes and better tuning Phedex agents can improve the staging from tape. Definitely starting to transfer the information in advance is a key item to take into account, as well as understanding the data distribution in the sites, data priority and deadlines.

T1_FR_CCIN2P3_MSS: 58 MB/s T1_FR_CCIN2P3_Disk: 1000 MB/s
Ticket: https://ggus.eu/?mode=ticket_info&ticket_id=127242 Tape

Got a pick rate of ~600 MB/s, and an avg rate of 300MB/s
- Buffer -> Disk Site admin reported rates ~1.5GB. Checking Phedex number numbers do not match. This might have to do with the fact that not only CMS transfers were taking place and not only production but also debug information was being transferred.

~260 MB/s debug + 295 MB/s prod (last 30 days) = 555 MB/s
~233 MB/s debug + ~250 MB/s prod (last 14 days) = 483 MB/

We conclude that the tests were successful, transfers from T0 Disk -> Disk endpoint gave and clear idea of how fast the data can be transferred to the site, since there were no target rates, understanding the link performance allows to be better plan the movement of data from T0 to T1 sites.
Regarding Buffer -> Disk transfers, tests were successful as well, since we also got and clear idea of the staging capabilities of the site.
Staging and data transfers can be optimized by serializing the files and enabling big tapes and better tuning Phedex agents can improve the staging from tape. Definitely starting to transfer the information in advance is a key item to take into account, as well as understanding the data distribution in the sites, data priority and deadlines.

T1_DE_KIT_MSS: 434 MB/s T1_DE_KIT_Disk: 868 MB/s
Ticket: https://ggus.eu/index.php?mode=ticket_info&ticket_id=127241 Tape monitoring: http://gridmon-kit.gridka.de/tapeview/cms/index.html

T1_ES_PIC_MSS: 175 MB/s T1_ES_PIC_Disk: 1000 MB/s
Ticket: https://ggus.eu/?mode=ticket_info&ticket_id=127238 Tape monitoring:

T1_RU_JINR_MSS: 116 MB/s T1_RU_JINR_Disk: 232 MB/s
Ticket: https://ggus.eu/?mode=ticket_info&ticket_id=127243 Tape monitoring: https://ggus.eu/index.php?mode=download&attid=ATT107344

T0 Disk -> JINR Disk (T0_CH_CERN_Disk to T1_RU_JINR_Disk graphic)

A peak of 240 MB/s can be reached and an average rate of 120 MB/s

T0 Disk -> JINR MSS (T0_CH_CERN_Disk to T1_RU_JINR_Buffer graphic)

Peak rate of 115 MB/s Average rate 65 MB/s

JINR MSS -> JINR DISK (T1_RU_JINR_Buffer to T1_RU_JINR_Disk graphic)

The staging test was performed twice

Based on Phedex plots, the rates from Buffer to Disk were:

  • 1st test -> 600 MB/s(peak rate) 250 MB/s (average).
  • 2nd test -> 600 MB/s (peak rate) 300 MB/s (average).

-- SiddharthMNarayanan - 2017-03-18

Edit | Attach | Watch | Print version | History: r27 < r26 < r25 < r24 < r23 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r27 - 2017-05-15 - JuanPulidoMojica
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    CMSPublic All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2020 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
or Ideas, requests, problems regarding TWiki? use Discourse or Send feedback