Conclusions of the face to face LSC-Virgo meeting
on data transfer for the S6/VSR2 run
Amsterdam-Nickef
24 September 2008
Participants:
- Oliver Bock (LSC)
- Leone Bosi (Virgo)
- Antonella Bozzi (EGO)
- Stefano Cortese (EGO)
- Scott Koranda (LSC)
- David McKechan (LSC)
- Livio Salconi (EGO)
Late:
- Bruce Allen (LSC)
- Stuart Anderson (LSC)
Data transfer tools
The bulk data transfer of h(t) and hrec data between LSC and Virgo will be performed using LDR in both directions.
In spite of the possibility of using the third-party transfer gridFTP capabilities EGO is willing to maintain the full featured LDR server in Cascina to benefit of the maximum flexibility and will communicate to Scott the supported OS version for the needed rpms.
Scott is willing to study a command interface to LDR to make easier the notification to LDR of file management actions on the storage at the endpoints (movement, deletion, etc.).
We forgot to clarify how will work the automatic file placement in Cascina for received files (and the data finding for files to transmit). As a minimum it is needed to change directory when a volume is full. Has Hannover the same problem?
Hrec off-line processed data transfers will be managed without an automatic procedure. They will be replicated manually using LDR once they are made available by the corresponding responsibles.
Hardware upgrades
EGO will upgrade the dataLDR server in Cascina and will ensure that the throughput with the storage will sustain the full LDR checksum.
The upgrade of the lnternet link at Cascina up to 500Mbps is ongoing.
Topology and latency
It is agreed that technically and administratively the best topology for the bulk transfers is that Hannover acts as the LSC-Virgo hub in Europe and therefore the LSC-Virgo data exchange could occur between Hannover and Cascina.
In addition Virgo has the requirement that the files to archive (due to Lyon Computing Center HPSS restrictions) have size not less than 1GB.
LSC is willing to provide a bigger version of Ligo files by adding frames up to the needed size before transmitting, and this “compacting” action can be performed at Hannover.
In order to overcome the delay introduced by the “compaction” buffering, an additional copy of data in the standard Ligo 128s frame files could be sent if Virgo needs so.
On the other hand LSC would need shorter in-time h(t) Virgo files respect to the ones transferred during S5/VSR1 (that were 3600s long) and asks what could be the size that could be provided by Virgo.
One other difference respect to the previous run is that the Ligo data stream will be continuous irrespective of the status of the interferometers (at least when there are data). Virgo should tell what will be the decision for the Virgo data stream.
Metadata
The simplest metadata needed to be exchanged to support the bulk transfers are the list of filenames to transferred in the given time period. These could be retrieved by querying directly the LDR database with an automatic procedure.
To support the analysis (either on-line or off-line) more metadata need to be exchanged at a higher level. A possibility proposed by Virgo is to setup a direct database connection between LSC and Virgo databases.
On-line data transfer and analysis
LSC is still evaluating whether LDR could be suitable for the real time data transfer for multi-detector on-line analysis respect to the dedicated NDAS tool.
Virgo is evaluating the use of direct Fd socket communication.
Next actions
Virgo has to:
- decide on what streams, file characteristics and latencies requests from LSC (and indirectly whether the moving of the LSC endpoint to Hannover fits)
- tell what minimal file length is willing to provide for in-time h(t) to LSC and whether the stream will be continuous or not
- study how and what metadata to exchange with LSC
LSC has to:
- provide LDR commands to facilitate the file management at the endpoints
- decide whether to setup a direct database connection for metadata exchange
EGO has to:
- upgrade the Cascina server and link and decide the official OS platform
Transfer test between Cascina and LSC
A data transfer stress test is foreseen for 27-31 October to exercise the network, the system and the storage at full speed.
The transfers will be performed manually and will not need the availability of the automatic procedures.
The following transfers will be performed simultaneously and singularly to test the maximum catch-up speed:
- the rawdata transfer Cascina -> Lyon
- the rawdata transfer Cascina -> Bologna
- the h(t) transfer Cascina -> LSC
- the h(t) transfer LSC -> Cascina