Transcript ispyb_status_2012x

Fall 2012 status & perspectives for the future…
ISPyB is made of…
JBOSS application
Heavily multi-threaded Java VM…
MySQL database
15-20 GB, still growing…
Files in a directory hierarchy
5-10 TB /data/pyarch, limited growth thanks to auto-archiving!
Quick history
pxweb
PC on ID14 (CC-0 ?)
no redundancy, no test box
gimmel
Dell 1750 in CR-106
no redundancy, no test box
pytest (Sun V120) in June 2007
March 2007
pyserv
Sun T1000 in CR-106
March 2008
pyserv
Sun T2000 in CR-106
recycling old T1000 as pytest
March 2012
pyprocz
Sun T5120 cluster
pydevcz moved to Sun cluster too
/data/pyarch created
moving MySQL databases to external server (myquickserv1/2)
Current setup
scud-mis1
scud-mis2
reliable
but single-room
sensitive on network issues
potential resource contention
does not scale well with autoproc (?)
reda
NetApp filer
Fully redundant
Located in CR-106
efficient dumps & cloning
but complex setup (no auto failback)
myquickserv1
myquickserv2
MySQL replication
Solaris cluster
Hosting:
 ispyb & ispytest
 mismwserv
 fmis2
Replica used for:
 Daily backup (dump)
 Test database (cloning)
In 2013 ?
Dell R720xd
Main characteristics:
 Dual Intel Xeon processors (e.g. E5-2640 2.5 GHz 6 cores)
 Up to 24 internal HDDs (16 TB high-speed storage)
 Ability to host massive RAM (64 GB by default) & 10 Gbit networks…
 Ability to add external disk trays for more storage
 Ability to configure more internal storage
 Not very expensive: around 15k EUR the box?
May become SC’s new beamline storage power horse !
Will be used for beamline local buffer in next months…
Possible setup
pyserv (1st R720xd)
Manual switch!
Based on logical IP’s – no client change…
pyspare (2nd R720xd)
Spare JBOSS server
Prod. JBOSS server
Mirrored deployments & config
Test JBOSS server
/data/pyarch
Prod. MySQL DB
/data/pyarch copy
Asynchronous – rsync based
Replicated MySQL DB
Test & dev. MySQL DB’s
Benefits & caveats
 All-in-one box:
 Dependencies are (much!) less complex – but there is a SPOF (MB)
 Manual failover triggered by users (BCU standbiers) is possible
 Faster access to database & files, hence a faster ISPyB (TBC…)
 Spare system:
 MySQL replication most usually fast (quasi-synchronous)
 Can we afford loosing latest additions in pyarch? How much?
 Manual failover:
 Robust to network problems (human decision to failover)
 Very easy to implement and maintain (MCS technique!)
 Bonus: potentially faster pyarch for beamlines!