On Wed, 14 Feb 2001 07:54:10 +0100, Bernie <[EMAIL PROTECTED]> wrote:
> Clarence wrote:
>> Bernie ?
> Yes?
>> Here is the lookahead stuff. I leave the block size at 1k and
>> set the lookahead to MAXIMUM (62k).
>> Speedchk says: 26,27 Sec. Arachne load Inbox 13 Sec. RELOAD INBOX 12 Sec.
>> We are allowed to set the lookahead to zero, so I did.
>> Speedchk says 24,24. Arachne loads inbox in 10 Sec. Reloads in 6 Sec.<g>
> I can understand a diffrence between the various sizes. This diffrence lies
> in the ammount of useless data that's in RAM, getting to the useful part
> would take a little longer if you need to to jump in RAM (this assumes that
> a cache in CPU and/or on the motherboard also uses read-ahead, let me know
> when this gets too complicated <G>).
Well certainly the cache in '586 cpu uses readahead, looking up both
possible outcomes of a branch instruction, but that is very different
from the HD cache which seemingly never has any intelligence involved.<g>
> Now the question is if all the 400 files in your mail directory end up in
> the cache, with a small sized cache (you only have ca 1.3MB for it) the
> time may increase.
> 1,3M/400 = 3 250 bytes/mail. Could you increase your cache size somewhat
> (so you get all of the files into the cache) and try again?
Sure thing ! But the most important thing for the cache to keep is the
directory structure and FAT so it knows where to look when it goes to
disk. BRB.
Ooops. I was gone so long I forgot the question. <g>
\MAIL 718 files ~ 3 Mb
*.cnm 403 files 1.8Mb, 2.2Mb occupied.
Cache size 2640k, empty. Block=2048 (minimum allowed for 2640k)
Lookahead = MAX = 62k. Load Inbox 11 Sec, Reload = 7 Sec.
That's different. Must be because the block size is different now.
So I tried the recommended block size of 8k. Load = 10, reload= 10.
Not much of a cache, eh ?
Now, put the block back to 2048 and lookahead to zero. Load=10, reload=6.
> However there are too many variables in use here for any of us to be able
> to give a clear answer. The things I can think of right now are:
> 1. CPU type
> 2. RAM type
> 3. Motherboard/CPU cache size and type
The above can be reduced to the generalizations that a fast computer will
be more help to the hard disk. And a slow computer needs all the help it
can get. <G>
> 4. HD cache size (and HD type, a SCSI drive will use less CPU - and so will
> UDMA drives)
> 5. Cache size
> 6. Read-ahead size
And the rest ? I'm working on it. ;-)
- Clarence Verge
- Still using Arachne V1.62 ....
