What sort of a chunk size have you set your cache-managers (ie, afs
clients) to use?
Michael Fagan <[EMAIL PROTECTED]> writes:
> Hi,
>
> We started running AFS over SL/IP for our home machines. The
> performance is tolerable for a 14400 line and will only improve as the
> speed of the line increases.
>
> The one observation I have made so far is that many things seem to
> travel over the (slow) link twice. For example, I read my mail from a
> POP server using (emacs) mh. When I type "inc" the mail travels from
> the POP server to my home machine. Then the mail is written to my
> home directory, which is in AFS, causing the *same* piece of data to
> travel back across the slow link to some AFS file server.
We've added a little feature in afs3.4 that may help alleviate this
slow response time. The client can be made to return immediately on
"close" of a file using "fs storebehind", so the copying back is done
asynchronously. Note however that now you cannot be sure that the file
did make it back safely (eg, if your quota on the volume was exceeded,
the async write-back will fail and the client is forced to drop the
file on the floor).
- Srivas
>
> The above is just one example of an application in a distributed
> environment that seems to assume fast(er) networking links.
Unfortunately, thats how most mail systems work. There's probably no
way you can prevent that "copy" from happening.
> Comments? How are others dealing with home access? What kind of
> problems are you having?
>
> Cheers,
>
> ----------------------------------------------------------
> Michael S. Fagan | IBM Research
> [EMAIL PROTECTED] | http://www.watson.ibm.com/~mfagan
> ----------------------------------------------------------
