Excutive summary: The old faster array makes a 3x difference in speed.
The copy was initiated on klausz, which mounts the /hsm from sam. This is the same fileset as was seen in greater detail earlier, and in fact both copies are visible in the week's worth of data.
Sam accepts the input and processes the user files into collections. Note the CPU time--far less is "wait" time than with the previous copy, both of which are displayed below. Thanks to a glitch in the nagios readout during reboots, I had to tinker with the zoom tool to get a view of the data, and the resulting images are smaller. (nagios gives 1TB/sec as a value during a failed read, which scales everything else into oblivion.)
Note that the disk I/O is 3x faster.
Chuma accepts the collections from sam and stores them. Below see CPU usage, disk read, disk write, and the total network activity.
I use the lustre /data/exp servers' disk read as a standin for contention for the network and for the disk service.
Modified 14-June-2011 at 08:49
http://icecube.wisc.edu/~jbellinger/StorHouse/14Jun2011
| Previous notes | Next notes | Main slide directory |