From: Martin Petder (martin@kungla.ee)
Date: Wed Oct 30 2002 - 10:35:14 EST
Hello all :)
On our ongoing effort of making some use of the massively buggy snapshot
feature of HSG80 we have another trouble to ask for experience :)
Namely we have observed severe reduction of the transfer speed from
snapshot unit after some normal-speed sequential read. At the same time
snapshot unit's regular 100% read I/O will change to 50% read, 50% write
I/O (as observable from VTDPY). Usually the speed drops from ~20MB/s
onto ~5MB/s in a few seconds and starts continous decline after that -
until both original and snapshot unit freezes (accepting no I/O from
servers), which is resolvable only by resetting the HSG80 controller
that the units reside on. Other units are not affected in any way (even
those on the same controller). VTDPY shows CPU highly idle even during
the hang.
This "feature" does not have any connection to the write I/Os to the
original or the snapshot units - it emerges even if there's no write I/O
at all for both.
So can anyone point some hindsight into the problem? How is it possible
to avoid the snapshot going 50% write 50% read (as if looping data with
itself)? :)
Waiting to summarize :)
The configuration:
MA8000, 2xHSG80, ACS8.7S, full cache, transparent failover, mirrored cache
2xSanSwitch 8EL switches (in same domain, interconnected)
Alphaserver ES40, Tru64 5.1 PK3, mounts original unit.
Alphaserver AS1000, Tru64 5.1 PK3, mounts snapshot unit and does backup
from it to MSL5026 via MDR using Veritas Netbackup Datacenter edition 4.5
-- Sincerely Martin Petder ======================== Kungla Dialoog C.P. tel. +372 6 115 300 fax +372 6 115 301 e-mail martin@kungla.ee ========================
This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:48:58 EDT