cache_dir's and NOVM

Greetings,

Just a few quick questions.. these have been bothering me for a while -
they may be stupid questions, but they bother me.

The excerpt below is from the Release-Notes v1.1..

*************
given:
        DS = amount of 'cache_swap' / number of 'cache_dir's
        OS = avg object size = 20k
        NO = objects per L2 directory = 256

calculate:
        L1 = number of L1 directories
        L2 = number of L2 directories

such that:
        L1 x L2 = DS / OS / NO
************

Now, should the last line be:

L1 x L2 = ( DS / OS ) / NO

or

L1 x L2 = DS / ( OS / NO )

?

I'm assuming that it is: L1 x L2 = ( DS / OS ) / NO

So if I had 4 * 6.2Gb cache_dir's with avg object size 20k, then a
suitable L1 x L2 combination would be:

16 x 128 = ( 6200 / 0.02 ) / 256

2048 = 1211

Is this approximately correct? I figured that the LHS being larger rather
than smaller is better. Or would it be better to use 16 x 96?

Question number 2:

This cache will run on a P166/Linux and will have 128M RAM to start with
- I'm considering running Squid-NOVM. The max requests per hour will be
~3000. Since the request numbers are small, I'm planning on using
ULTRA-DMA/33 EIDE drives - not SCSI - and the machine will only be running
Squid.

Filesystem blocksize will be 8k. And as stated before, there will be 4 *
6.2gb cache_dirs.

Correct me if I'm being irrelevent, but gnumalloc will be used (is
dlmalloc better, or have I lost the plot?).

Is the above a healthy way to go? Have I overlooked something?

Any advice and comments on the suitability of NOVM and the figures above
would be greatly appreciated.

Thanks in advance..

Umar.

--
Umar Goldeli
SYNFLUX International P/L.
P.O. Box 98, Five Dock, N.S.W. 2046, Australia
e-mail: umar@synflux.com.au 		Phone: +612-9712-2411
WWW   : http://www.synflux.com.au	Fax  : +612-9712-2399