PM>> with a 2MB cache and lazy write turned on. The disk drive with the

 PM>> message base is a Maxtor 540MB with 8.5 ms access time.



 db> ..and you have at least 16Mb RAM, right?  That accounts for the 2+Mb

 db> memory that Squish is reporting, and explains why your throughput is so

 db> much faster than mine.



Yeah, I've got 20MB, although I've got an ISA SCSI controller, so the upper
4MB is just swap space.



 db> Only i486DX 33MHz, message database on HPFS partition, on a Conner

 db> CP30104 116Mb IDE driver, with an access time of around 11ms (give or

 db> take a few). Cache settings are:



 db>           DiskIdle:    65000 milliseconds

 db>             MaxAge:    65000 milliseconds

 db>         BufferIdle:    16000 milliseconds

 db>         Cache size:      384 kbytes

 db> Lazy writes are enabled.



I don't suppose there's any way to get this info for a FAT cache is there?



 db> Machine only has 8Mb RAM, and has occasion to swap when running multiple

 db> compilers (along with the other mishmash running in the background).  The

 db> best throughput that I have still recorded in my logs is 12/s (it's been

 db> higher, but not often).



Multiple compilers? At once? I'm impressed :-)



 PM>> If you use the second approach, then you have to run SQPACK to delete

 PM>> all the old records. It also packs the database to remove the unused

 PM>> space. I usually run it every couple of days, but you can run it as

 PM>> often as you like. On a large message base it could take a while though.



 db> I run it in the background from a cron job at 1:00am every morning

 db> (assuming I'm using the computer then, which is a pretty safe bet);

 db> generally doesn't take too long - but then again, when you're not

 db> watching it you don't really care.



I don't mean it'll take 30 minutes or anything like that. My message base
is sitting at about 8MB and it takes 35-40 seconds to run SQPACK. So on a
larger message base it'll take a few minutes.



 PM>> There's a parameter in the Squish configuration file called 'buffers'.

 PM>> If you haven't already, make it 'buffers large' and see if it helps.



 db> Already done.



Oh well, that's the extent of my Squish tuning knowledge, so I'm out of ideas.





Paul



--- GoldED/2 2.42.G1114