The Question is:
Performance problem in two nodes cluster. Nodes identical in hardware/software
configuration and shared storage. In one node an executable running from DCL
with take three minutes while in the other it takes 40 minutes. It is the
same executable using t
he same files. I have been able to identify the problem as the process going
into MWAIT state (RWSCS).
What is causing it and how do I fix it? Thanks.
The Answer is :
RWSCS is a resource wait for system communications services and implies
that there are cluster communications or configuration problems lurking,
or that the cluster is encountering excessive I/O or locking activities,
credit waits or other similar overloading.
You mention shared storage, but not the specific details of the storage
technology. Shared SCSI still depends on the network for locking, as
Use the network counters (LANCP SHOW DEVICE /COUNTERS, etc) to check
for network-level communications errors, assuming an Ethernet or IEEE
802.3 communications network is used.
Use the DCL command SHOW CLUSTER/CONTINUOUS and ADD CR_WAITS to see
the cluster credit wait activity.
Use MONITOR and other tools to examine disk activity. MONITOR DISK
with /ITEM=QUEUE_LENGTH can be used to detect overloaded spindles.
Sustained queue depths of 0.5 or larger are usually considered to
indicate an I/O overloaded; to be bad.
The OpenVMS Wizard will assume that you are aware that OpenVMS Alpha
V7.1-2 is no longer supported, and that an upgrade to a more recent and
particularly to a supported OpenVMS release is recommended. (Please
see the "PVS" links in the FAQ for support information and scheduling.)
You will want to apply the mandatory ECO kits for the release in use,
whatever release that might be. (The FAQ has pointers to an ECO search
engine; to a tool which can identify ECO kits by installation rating.)
More recent OpenVMS releases have various lock manager and I/O
optimizations, updates that greatly improve RMS file locking (V7.2-1H1
and later), new and larger I/O caching (V7.3 and later), as well as
numerous other performance improvements.