SUMMARY : Size/Performance of CacheFS [Follow Up]

Chi Man Wu - SAN (wuc@hkmail.sps.mot.com)
Wed, 26 Mar 1997 15:52:50 +0000 (HKT)

Hello admins,

Special thanks to all for more detail follow-up:
Leif Hedstrom leif@netscape.com
Paul T. Keener keener@upenn5.hep.upenn.edu
David Harel. hareldv@netvision.net.il

Best regards,
Man Wu

Original question :
>
> Hi,
>
> We want to setup CacheFS of our system (Solaris 2.5.1),
> My question is,
>
> 1) Is there any recommendation ?
> 2) How to monitior the performance ?, i.e I can fine tune
> the size of it ?
>
> I will summarize...
>
> Best regards,
> Man Wu

--------------
>From Asim Zuberi <asim@psa.pencom.com>

Your best bet to check out the white paper at Sun Site.

http://www.sun.com

search cacheFS

--------------
>From cda <cda@cosmoslink.net

F.Y.I.
It is my understanding that you cannot have cachefs and automounting. I
know this is true for Sol 2.4 and have heard it is for 2.5.1.

--------------
>From Cheng, Bruce <Bruce.Cheng@Aspect.com>

The following two commands can tell you how efficient cachefs is being
utilized.

cachefslog cachefslog (1m) - Cache File System logging
cachefsstat cachefsstat (1m) - Cache File System statistics

--------------
>From Eirh-Yu Hsie <hsie@al.noaa.gov>

http://www.sun.com/sunworldonline/swol-03-1997/swol-03-perf.html

#################################################################
# Follow-UP Follow-Up Follow-Up #
#################################################################
--------------
>From Leif Hedstrom <leif@netscape.com>

F.Y.I: The above statement is not true. cachefs and automount works just
fine together. I use it all the time.

--------------
>From Paul T. Keener <keener@upenn5.hep.upenn.edu>

It is possible to cache an automounted filesystem; in fact, it is documented
in the man pages. However, what is not documented is that the performance
is *really* bad, at least for 2.5.1. Once the filesystem is mounted, there
is not problem, but mounted the filesystem can take a long time (perhaps
as long as a minute). I have not seen any cause for this behavior, not
could I find anything to tune to avoid it.

--------------
>From David Harel <hareldv@netvision.net.il>

First I tried it on Solaris 2.5 and I used autofs (automount maps). It
works !!

I wanted to create a case in which all the data needed by the client
will eventually reside in the cache directory and thus reduce network
traffic. So I defined very big cache directories ( about 600MB )

I am sorry to tell I didn't get a much better performance from the
client ( user hardly felt any different ). Moreover in some cases the
cache directory got jammed some how and the applications began doing all
sort of funny stuff until I flushed the cache and rebuild it again.

Quite a good explanation I got from a sales men ( not very reliable ? )
of network appliances. He claimed that most of the time the system is
waiting for the disk to perform seek therefore the cachefs which used
intensive disk operation didn't help removing the bottle neck. Of course
he had better solutions, using their special NFS server which overcomes
the seek problem.