Original Question :
I am currently in the process of upgrading our Unify and Oracle production
database server from a AlphaServer 8200 5/440 to a Compaq AlphaServer GS60E
6/700 and have noticed that /sbin/update enters an abnormal state where it
consumes excessive amounts of CPU time following the return from periods of
light to moderate system usage (update consumes up to 5 seconds every
minute). As a result of exhaustive testing I have been able to reliably
reproduce this bad behaviour by just running consecutive "setld -i"
commands !! Digital Unix 4.0F patch kit 3 was initially installed on both
servers however the GS60E has since been upgraded to patch kit 4. A SWXCR
RAID controller card is employed to provide mirroring for our root, swap
and usr filesystems. Application filesystems are supported by dual HSG80
controllers connected via a Compaq StorageWorks SAN switch to a PCI Fibre
Channel adaptor card KGPSA-BC (Driver Rev 1.22). This odd behaviour
continued, but to a lesser extent, when the SWXCR raid controller was
replaced with a standard SCSI controller and single system disk and after
the KGPSA-BC card was removed.
Summary :
This misbehaviour was discovered to be due to a problem with the AdvFS
hashing table on 4.0X. Compaq has provided a patch which has rectified this
problem. Administrators experiencing similar problems should contact their
local Compaq representative.
Received on Thu Oct 12 2000 - 01:03:44 NZDT