File System Cache
Tuning


Mr Cameron,

I hope you don't mind me emailing you but I have a quick question about your VMS tuning hints page. You mention that the ACP_EXTLIMIT parameter should be 50% larger than the ACP_EXTCHACHE parameter. My problem is that within SYSGEN they are in different units. The EXTLIMIT is in PERCENT (of what?) and the EXTCHACHE is in EXTENTS!

Anyway if you've gone this far into the email without binning it you may be interested in what / who its for. The system is an airport flight information system that has been written in C under VMS version 6.2 on two alpha server 4000's. The clients (250) connect using netbios over TCP/IP to pathworks drives. Basically we are having problems with TCP/IP connections dropping and are trying to find the correct tuning set up for our site.

I hope you can shed some light on the ACP_EXT units problem so thank you in anticipation for you time

Best Regards
Jez Ross-Clyne

Airport Systems Group
EDS Ltd, Hartley House 15 Bartley Wood Business Park,
Bartley Way, Hook, Hampshire, RG27 9XA
Tel: 01256 741308, Fax: 01256 741111

Email: jeremy.rossclyne@transport.uk.eds.com

Jeremy ...

Thanks for the question.

That's a very good question you brought up, and I'm going to amend my page to make it more clear, since they are indeed different units. The value for ACP_EXTLIMIT should be at least %50 larger than the value for ACP_EXTCACHE. The parameter ACP_EXTLIMIT is in thousandths of unallocated disk blocks, and ACP_EXTCACHE is indeed in extents (number of pre-allocated extents that have been allocated on the disk and are ready to be given to files requesting an extension to a file).  The %50 rule is only a guideline, but I heavily recommend it if you ever have more than 3 or 4 files open for write at one time on any disk. However there is a tradeoff. Making ACP_EXTLIMIT too big in comparison to ACP_EXTCACHE can cause your files to become more fragmented when multiple files are being written to at the same time.

File fragmentation and multi-header files can slow your disk access tremendously. For this, you can dismount your disk from the cluster, do an image backup and restore to defragment the files. If this method is unreasonable because you need to keep the data online, I highly recommend you get a disk defragmenter. I recommend DISKEEPER by Executive software.

Since this is a server with many clients, it is important for you to constantly monitor your file cache during peak periods, as mentioned in my web page.

However it is very-very unlikely that this is causing your TCP/IP connections to drop. For this problem, I would look at the counters for TCP/IP to see where your problem lies, and then go from there. Make sure you read the installation documentation for the TCP/IP product you use, paying close attention to any mention of parameters. The ACP_ parameters will have little if no impact on networking problems, just speed of serving files over the net.

Hope this helps.
Jeff Cameron


Send me your question. 


 

My Home Page | VMS Home

DCL | Utilities | Management | Tips

FORTRAN | Pascal

eMail Questions

Quiz?