On a Linux server
sar -b reports much too high read block statistics, sometimes 3.500.000 readb/s peak and around 2.500.000 readb/s sustain, which makes more that 1 gigabyte/s (according to the sar man page, on 2.6 kernel 1 block = 512 bytes).
This is a 4 year old server running SLES10, with san atachement. The SAN storage hardware is unlikely able to give such volume of data per seconds.
When I do
sar -d I see dm-* devices individual statistics, some devices are SAN luns and some are logical volumes. As the logical volumes are build with physical luns, is it possible that the global read-block/s statistic of
sar -b is counting the I/O twice ?
I must precise that it seems not to be due a faulty card, that I/0 scheme follows a schedule, and has a duration and goes to 0 in idle time