Re: Much wanted feature.

On Tue 28 Oct, 1997, Stewart Forster <slf@connect.com.au> wrote:
>> As for mounting a cache_dir remotly, I think others have had only bad luck
>> with this. My guess is that it would be very slow, although I havn't ever
>> tried it.

We've tested such things with back-to-back 100BT, to a NetApp
filer. Worked quite well. The key thing is that all the seeks to find
data are embodied in just one NFSop, so all that work is offloaded
onto your NFS server. Additionally, all writes can be consumed
instantly (if you've got NVRAM or you're using NFSv3) by the server.

We also experimented with a RAID-3 controller with large amounts of
cache. That also worked very well for Squid and we didn't manage to
stress it at all (disappointingly.)

> Not to mention all the NFS locking required. Yeech.

Well, you don't need any NFS locking because nothing else is sharing
the cache.

-- jrg.
Received on Tue Oct 28 1997 - 03:21:52 MST