Re: [squid-users] disk partition locations ?

On Fri, 9 Sep 2005, Matus UHLAR - fantomas wrote:

> There are expectations that at the begin of disk there should be lower seek
> performance, but better performance for sequential reads, and in the middle
> of disk there should be higher seek performance wit lower performance for
> sequential reads.

The data rate of modern drives is about the same across the disk as they
use progressive cylinder sizes to make use of the increased lengths of the
outer tracks. The "Cylinder/Head/Sector" notion is purely virtual these
days, mainly there to keep PC BIOS & DOS happy (the rest of the world
does not care)..

Seek times is also close to the same across the disk, mainly dependent on
the length of the seek (next track seeks very fast, longer seeks a bit
slower). But with the outer tracks being bigger, slightly less seeks may
be needed for the same amount of data..

> However ad I use two whole drives on all of my servers, I haven't checked
> this.

My preferred setup for a reasoably performing Squid is three drives

1 drive for OS and logs (incl swap.state),. Maybe mirrored for
redundancy (no noticeable performance difference). (if mirrored then 4
drives is used in total).

2 drives for cache, one aufs cache_dir each. ext2 filesystem with noatime.

In smaller systems where performance is not the most important I use two
drives like

1. A software mirror partition of the OS + logs. ext3 filesystem.

2. The rest of each drive used for cache, one cache_dir each. ext2
filesystem with noatime.

And in really small systems a single drive is used. (or even none,
depending on the purpose).

It is very important to have sufficient amount of memory in your Squid
server.

In the past I have experimented with using reiserfs for the cache, and
while it provided slighlty better performance results than ext2 (provided
not short of memory) it did not provide the same level of robustness in
case of media failure.

the ext2 cache filesystems is automatically wiped clean on boot if fsck
complains on anything. While the cache partitions is fsck:ed after a
unscheduled restart of the server a non-caching Squid is running to take
care of the service..

Regards
Henrik
Received on Sun Sep 11 2005 - 06:49:23 MDT