|
HP OpenVMS Systemsask the wizard |
|
The Question is: Is there any way to minimize rms locking, besides opening files for exclusive and/or read only access? We have some multiprocessor sites that are seeing very high mp sync times (250% of 6 cpus) at times. We were able to make some files read only and tha t helped but these companies run tons of reports on transaction files and its kills the systems. These are 8400's with 625mz cpus. By the way the question about the product RXAXP is Raxco's Performance monitoring suite (PerfectCache and PerfectDisk) Their license is not a traditional VMS license. You can list their licenses by running licifno.com in the rx_v*: directory. Show log rx* to see the exact directory. Thanks in advance for your time. The Answer is :
There is presently no way to specifically minimize locking, save for
those techniques you are already aware of.
OpenVMS Engineering has begun investigation of an enhancement that
would allow "readers" to override any record locking in accessing an
indexed file that is opened for shared write access.
Presently, for the purposes of generating "tons of reports" -- and
assuming the reports can lag behind the current transactions by a few
hours -- a read-only (temporary) copy of the database could be created
for reporting purposes. The copy could obviously be duplicate the
indexed structure of the existing file should keyed access be required
for the reports, or it could be a converted sequential file should that
format be sufficient for the purpose.
Various RMS global buffer enhancements were introduced in OpenVMS
V7.2 that improve performance on shared write files, and these have
been backported and will be available on OpenVMS V7.1 (VAX and Alpha).
These enhancements provide:
o A new algorithm for global buffer management that dramatically
improves scalability. The performance associated with the
previous algorithm effectively limited the maximum number of
global buffers permitted on large, shared files. With this
change, the number of global buffers can be increased to the
full limit of 32,767, to exploit large memory systems.
o New synchronization to the global section used to implement
RMS global buffers. This synchronization uses inline atomic
instruction sequences, rather than distributive locking.
This change permits increased concurrent access to the RMS
data in the section, particularly on symmetric multiprocessing
machines (SMP).
The first enhancement benefits thoses who wish to use very large
RMS global buffer counts. The second helps any application using
RMS global buffers where contention on the global section data is
the bottleneck.
|