Re: diskstriping or muliple cachedirs

On Tue, 16 Sep 1997, James R Grinter wrote:

> If you are going to Raid-0 stripe disks, your stripe/interlace size
> (different names with each OS, same concepts) needs to be picked with some
> knowledge of the OS and filesystem in mind. With Solaris, you also want to
> build the filesystem with a suitable maxcontig value (chunk size =
> maxcontig/spindles,).
>
> A book I have (Configuration and Capacity Planning for Solaris Servers)
> recommends that maxcontig and the stripe width should be some multiple of 32kB
> (maxcontig defaults to 56kB), and goes into a lot more detail than I could
> summarise here. It also points out that for random access (which is pretty much
> what we get, though we personally see 2/3 write to 1/3 read the majority
> of the time: that's with the VM Squid) changing maxcontig isn't critical.

Sorry if this becomes a Solaris-centric thread, but:

Has anyone evaluated using fastfs vs. logging filesystem ?
Can anyone comment on issuing 'sync's every few seconds ? How many seconds
would be a good value ? I have seen running a syncloop greatly increase
CPU usage :-(

Using a syncloop is said to increase throughput by not hitting the disks
too hard when a sync occurs every 30 seconds (default).
You need to have some spare RAM for disk buffering, though.

BTW, try setting cache_swap_low equal to cache_swap_high, this results in
equal disk usage over time as opposed to a big hit when cache_swap_high is
reached.

BTW, Adrian Cockcroft, tuning-guru at Sun, recommends a 128K stripe width.

>
> James.

Markus
Received on Wed Sep 17 1997 - 08:50:20 MDT