[comp.unix.ultrix] Will prestoserve help a client

puglia@cunixf.cc.columbia.edu (Paul Puglia) (03/12/91)

Will prestoserve help improve NFS performance if it is 
installed on a client?  My gut reaction is no, but  
DEC presales claims that it might.  Who is right?


Paul Puglia
Columbia University
Department of Civil Engineering

puglia@cunixf.cc.columbia.edu

mjr@hussar.dco.dec.com (Marcus J. Ranum) (03/12/91)

puglia@cunixf.cc.columbia.edu writes:
>Will prestoserve help improve NFS performance if it is 
>installed on a client?  My gut reaction is no, but  
>DEC presales claims that it might.  Who is right?

	Nope. It might improve other client file-system I/O (writing file
system meta-data) but it won't help the client do NFS.

mjr.

pcg@test.aber.ac.uk (Piercarlo Antonio Grandi) (03/18/91)

On 11 Mar 91 21:38:00 GMT, mjr@hussar.dco.dec.com (Marcus J. Ranum) said:

mjr> puglia@cunixf.cc.columbia.edu writes:

puglia> Will prestoserve help improve NFS performance if it is 
puglia> installed on a client?  My gut reaction is no, but  
puglia> DEC presales claims that it might.  Who is right?

mjr> Nope. It might improve other client file-system I/O (writing file
mjr> system meta-data) but it won't help the client do NFS.

Rethorical question: how many megabytes of extra disk buffer cache on
the client or server can you buy for the cost of one NFS accelerator?
--
Piercarlo Grandi                   | ARPA: pcg%uk.ac.aber@nsfnet-relay.ac.uk
Dept of CS, UCW Aberystwyth        | UUCP: ...!mcsun!ukc!aber-cs!pcg
Penglais, Aberystwyth SY23 3BZ, UK | INET: pcg@aber.ac.uk

p554mve@mpirbn.mpifr-bonn.mpg.de (Michael van Elst) (03/18/91)

In article <PCG.91Mar17180404@aberdb.test.aber.ac.uk> pcg@test.aber.ac.uk (Piercarlo Antonio Grandi) writes:
>Rethorical question: how many megabytes of extra disk buffer cache on
>the client or server can you buy for the cost of one NFS accelerator?

Lots and lots but on the server the NFS accelerator isn't just an extension
to the buffer cache. It is an extension of the disk drive that makes
synchronous NFS operations much faster.
On the client it doesn't help and you'd better go with more memory for the
regular cache.

Regards,
-- 
Michael van Elst
UUCP:     universe!local-cluster!milky-way!sol!earth!uunet!unido!mpirbn!p554mve
Internet: p554mve@mpirbn.mpifr-bonn.mpg.de
                                "A potential Snark may lurk in every tree."

mjr@hussar.dco.dec.com (Marcus J. Ranum) (03/18/91)

pcg@test.aber.ac.uk (Piercarlo Antonio Grandi) writes:

>Rethorical question: how many megabytes of extra disk buffer cache on
>the client or server can you buy for the cost of one NFS accelerator?

Straight-man answer:
	Since the bottleneck is usually writes, especially with NFS,
added cache in the client won't help much if you do any writes (unless
you're paging a lot, in which case paging == writes, and adding memory
will reduce writes). Since writes on the server are also synchronous,
adding extra disk buffer cache will only improve your read times, not
your write times.

Rhetorical answer:
	How much local disk can you buy for the cost of all that memory?
If performance is your problem, and you're doing writes, NFS should probably
not be part of your solution.


mjr.
-- 

	The world is just backing store for virtual reality games.

frank@croton.nyo.dec.com (Frank Wortner) (03/18/91)

In article <PCG.91Mar17180404@aberdb.test.aber.ac.uk>,
pcg@test.aber.ac.uk (Piercarlo Antonio Grandi) writes:
 
> Rethorical question: how many megabytes of extra disk buffer cache on
> the client or server can you buy for the cost of one NFS accelerator?

Buffer cache really doesn't fulfill the same function.  Unless the cache
is nonvolatile,
a system crash will destroy any data that hasn't been flushed to disk. 
Since PrestoServe
commits transactions to "stable storage," it can safely buffer
transactions that normally
can only use a buffer cache as a "write-through" cache, if they can use
it at all.

The other assumption you're making is that this is all done by expensive
memory.  If that
were the case, it should be possible to simply increase the size of the
buffer cache and
watch disk performance go through the roof.  If that was all, then
someone would have
surely discovered this by now.  NFS accellerators also use sophisticated
software to manage
and control the cache.  The cache itself is only half the story.  The
hardware/software
combination is what makes the product.


						Frank