Monthly ArchiveMarch 2005
Dcache & Work admin on 15 Mar 2005
15/03/2005
Fixed read problem by restarting srm door.
Patched pnfs2ads2 script to accept pnfs parameter to allow it to be reused without changing the code.
Disabled firewall on gftp0441 after discovered it was still running.
14/03/2005
Investigating why unable to read from production dcache dteam pool, when we can write tothe pool and read from other pools
Reinstalled gftp0441.gridpp.l.ac.uk after new motherboard was fitted by Compusys. Things not in kickstart are: setting up of second interface, adding rl.ac.uk link into /pnfs directory and turning off firewall.
Removed fetch-crl crons from SC machines that don’t need certificates.
11/03/2005
Setup CA people on our CVS system as a temporary expedient until a new ESC CVS server is in place. Got a reponse to a mail I’d sent to the dcache developers asking about transfer rates pointing us at files under /opt/d-cache/billing, spent some time with Dave Kant decoding them and showing him around the production head node, appears he may be able to get what he needs from the postgres database.
10/03/2005
GPC Training in morning, TOAST meeting in afternoon
Began setting up csfnfs60’s first pool to do ads transfers.
09/03/2005
Analyzed results of test - jobs finished after 12 hours due to proxy expiration, pool emptying did work but network rate was reduced from 3Gb/s to 2 GB/s and load on disk servers increased, however CERN will handle deletion from their end so removed empty pools script from cron
Dcache & Work admin on 08 Mar 2005
08/04/2004
Trying a new method of emptying pools - in the pool admin interface type pool lfs volatile. This seems to set the mode of files stored into this pool as cached and not precious, meaning they will get deleted if space is needed. Running 40 jobs to fill SC dcache to confirm this.
Got DNS entries for CERN-facing interfaces, but it looks like they won’t be used - we’ll have a hosts file on each machine at CERN.
07/04/2004
Investigated why jobs over weekend copying files into dCache SC stopped - turned out that we’d filled the disks. 3.5Gb/s fills 12 TB in 7-8 hours. Implemented a cleaner script using expect that instructs the pools to empty themselves - it doesn’t inform pnfs that the files are no longer there however, could be problematic if someone wants to read a file back…
Discussed with various people the reason for the 3.5Gb/s apparent maximum - increasing the num ber of active puts in the srm did little. Eventually decided that the disk servers’ CPU are probably the limiting factor.
Added cron job to both dcache’s postgres hosts to vacuum the database every 6 hours. Full vacuums don’t seem to work(hang) possibly because the dcache srm connections to the database never seem to go away, so not performing a full vacuum automatically for the moment
04/04/2004
Inducted into the ways of edg-job-submit by Steve, ran 10 jobs against SC dCache to see what performance we could get over the weekend, 10 jobs managed to get 1Gb/s, Steve tried on Saturday with 30 jobs and managed to get 3.5Gb/s, seeing as the UKLight link to CERN if it ever works is 2Gb/s, it looks like the infrastructure will be fine.
Dcache & Work admin on 04 Mar 2005
03/02/2005
Got information about IP addresses for second initerfaces so configured systems to use these, gftp0441 hung bringing up the eth0 so asked George to take a look at it.
Did a mass reboot of the gridftp nodes and noticed later that they’d failed to connect to the head node, doing them individually seemed to work though - suspect it was too many at once - worth bearing in mind for the future though.
Dcache & Work admin on 02 Mar 2005
Completed installation of gridftp servers , now able to do srm get and put from lcgui01 successfuly into service challenge dcache. Automated script is good but some things still need to be done by hand - installation of certificates, setting up of symlinks (although this should be automatable but for some reason the head node is using rl.ac.uk instead of gridpp.rl.ac.uk as its pnfs domain).
Have now gone as far as I can without details of how we connect to CERN
Dcache & Work admin on 01 Mar 2005
01/03/2005
Wrote script to install dcache gridftp server in push button mode - surprisingly little to it, usual fun with quoting brackets in regexps, had to quote all the $’s in a string to stop variable interpolation too.
28/02/2005
Continued configuring srm node for service challenge, discovered that the problem I had with network configuration were due to a script further along hard coding the values to eth0, replacing that with one for eth1 solved the problems
Began installing gftp node