For a ’service grid’ guy, it is quite interesting to see the major problems of the hardcore HPC community. The most funny thing is that our Grid-Occam motives are not too far from reality:
Archive for March, 2006
SOAP vs. SOA
March 20th, 2006 by PeterSome nice try to give a clarification of the fuzzy web service discussion – “SOAP and HTML” vs. “SOA”:
http://www.optimizemag.com/article/showArticle.jhtml?articleId=180207087
WSRF vs. WS-Management
March 20th, 2006 by PeterFrom GridToday:
HP, IBM, Intel and Microsoft announced on March 15 their intention to work together to “develop a common set of [Web Services] specifications for resources, events and management that can be broadly supported across multiple platforms.”
They also published a roadmap outlining what these specifications will be.
The published roadmap suggests that the new specifications that are to be developed will include essentially all of the core concepts introduced back in 2001 in the Open Grid Services Infrastructure (OGSI) and subsequently incorporated in WSRF/WS-N.
Sliced bread
March 14th, 2006 by PeterDuring several DRMAA phone conferences, I learned how american guys express a sarcastic praise – “The best thing since sliced bread”.
It is funny enough that Wikipedia has an own article. However, we also found a nice comparison between the iPod and sliced bread.
KISS Backup
March 12th, 2006 by PeterBackup procedures, either at home or company, are hard to establish and hard to maintain. During the last years, I experienced that working solutions must be fully automated. If you need any kind of periodical interaction, you are lost.
The second major experience within the HPI and the ASG project is that spending money for backup solutions does not ease your work. Commercial backup software is complicated, buggy and never fits to your particular use case. For this reason, we reverted (after several iterations) to traditional the old-fashioned dump / restore tools and a tape drive in ASG. A cron job does a daily dump of the harddisk and pipes the result with “rmt” to the machine with the streamer hardware. Other (more critical) data is copied hourly with rsync to a remote machine.
Lately, we had a harddisk crash and a real use case for full restore. In fact, it needed only a combination of 3 Unix tools (and 1 hour) to perform a remote restore from the machine with the tape drive to the particular server. The buffer tool is needed to ensure a continuous data stream from the streamer, and we prepared the streamer tape position with the MT tool:
ssh root@streamerpc "buffer w -p 75 -m 5M -i /dev/tape"|restore
Therefore, I suggest sticking with the KISS principle – “keeping it simple and stupid”.