Checksumming support in SRM implementations

Some applications have requested the availability of checksumming support on files stored on the storage elements in the Netherlands. The first VO to request this was Atlas, but other VOs are also very interested. Unfortunately, checksumming support seems to be non-homogenously implemented across the different storage backend systems.

This Wiki page is the result of a short investigation into the checksumming differences between dCache, DPM, StoRM and CASTOR.

To store a file with a checksum in SRM I've used the commands

lcg-cr -d $SRM -l lfn:/grid/pvier/janjust/myfile file://$PWD/myfile --checksum
lcg-cr -d $SRM -l lfn:/grid/pvier/janjust/myfile file://$PWD/myfile --checksum --checksum-type adler32
lcg-cr -d $SRM -l lfn:/grid/pvier/janjust/myfile file://$PWD/myfile --checksum --checksum-type md5

where $SRM points to the appropriate storage system.

Note: so-called gridftp checksums are checksums that can be calculated on-the-fly using a special GridFTP command

dCache

• srmping: v2.2 dCache production-1.9.3-3
• seems to support only adler32 checksums (which is what WLCG seems to want) ; according to Ron md5 is also support but this requires a server reconfiguration
• checksums
  • are computed when the file is transferred to dCache or during a move between pool nodes.
  • are stored in the dCache namespace
  • can be retrieved using the srmls -l command
• gridftp checksums are not supported

DPM

• srmping: v2.2 DPM 1.7.0-5
• supports adler32, md5 and crc32 checksums
• supports only gridftp checksums; adler32 and crc32 are supported using a special DPM-DSI GridFTP plugin
• checksums
  • can be computed using the lcg-get-checksum command
  • are stored in the DPM namespace but can never be retrieved ( !! )
  • are not displayed when using the srmls -l command
  • are recalculated every time

StoRM

• srmping: v2.2 StoRM <FE:1.4.0-01.sl4><BE:1.4.0-00>
• supports only md5 gridftp checksums
• checksums
  • can be computed using the lcg-get-checksum command; however, the command fails for large files
  • are not displayed when using the srmls -l command
  • are recalculated every time

CASTOR

• srmping: v2.2 CASTOR v2_7_15 2.1.7
• supports only adler32 checksums
• checksums
  • are computed in the background
  • cannot be specified with the --checksum option
  • are displayed when using the srmls -l command
  • are stored in the CASTOR name space