In Squid's mailing list, on Sat, 27 Jul 1996, Ong Beng Hui wrote:
> > well, i've been testing with 34GB of disk, but that is only testing.
> > Mainly trying to get a feel for the best combination of stripes/multiple
> > cache directories. The striping is winning, and with the modifications
> > for cache dir structure in 1.1 would be better yet.
> 
> Do you have any stats on that ? Like for instance, 
> how much faster, how do you go about stripping the
> disks ?
> 
> I am using Harvest Cache V2 and I am using a DEC 4100 
> here with 3 channel RAID card. I am trying to test
> the performance without stripping. 
> 
> Our production machine here is also a DEC 4100 and
> we strip 3 disks with just 2 SCSI channel. Performance
> is okay with 1 million plus hit per day. 
The options are:
1.   1.0.x's 100 dir  or  1.1.x's 16*256 dir
2.   single           or  multiple cache partitions
3.   JBOD             or  RAID0
3. if jbod, there's no guarantee requests will not converge onto a
   single disk, even if you use multiple cache dir and 1.1's 16*256
   dir. => you've _NO_ control which disks requests land on. you would
   like stripping on the (scsi) block level, not on a per-file (JBOD)
   basis.
2. if RAID0 is the answer to #3, you'll want single cache partition.
1. regardless of what you choose for #2, #3, many files on a single
   dir is going to slow down your OS.
hence,
1. 1.1.x's 16*256 dir
2. single cache partition
3. RAID0
btw, your production machine is (3|4|5)0 GB big and only 3 disks?
each disk must be 9GB huge. RAID0 won't like these beasts, it
prefers many small ones.
--- Ivan Leong <ivanl@pacific.net.sg> the power to move Systems Engineer, Pacific Internet http://ilsp.pacific.net.sgReceived on Sat Jul 27 1996 - 01:10:09 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:32:42 MST