home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!gatech!prism!gt1111a
- From: gt1111a@prism.gatech.EDU (Vincent Fox)
- Newsgroups: comp.sys.sun.hardware
- Subject: Re: Prestoserv?
- Message-ID: <66105@hydra.gatech.EDU>
- Date: 16 Aug 92 07:01:52 GMT
- References: <1992Aug4.104143.136@csc.canterbury.ac.nz> <1992Aug14.042821.18360@cssc-syd.tansu.com.au>
- Distribution: usa
- Organization: Georgia Institute of Technology
- Lines: 43
-
- In <1992Aug14.042821.18360@cssc-syd.tansu.com.au> rodney@cssc-syd.tansu.com.au (Rodney Campbell) writes:
- >adm@einstein.canterbury.ac.nz (Andrew McGregor) writes:
- ]]We saw this gadget mentioned in sun's pricelist.
- ]]some questions:
- ]]I gather it speeds up performance for server type activities;
- ]]by how much? Exactly which activities?
-
- >It basically speeds up NFS writes by caching incoming writes over NFS and
- >then writing them out to disk when there is time. You put the card and
- >software on the server and the NFS clients don't know the difference.
- >It does make a difference (only to NFS writes) but how much I don't know -
- >I have heard figures such as 600% quoted.
-
- For moderately sized sequential writes it's the greatest thing
- since sliced bread. Look at nfsstat -s and see if your write percentage
- rate is much above 10% to see if this is going to help much.
- If you've got I/O-bound applications with moderate to large sequential
- writes, you might want it. I'ts pretty expensive, at around $2K.
- In our CAD environment, writing model files of 512K to 2M size is
- pretty common, making this baord very desirable.
-
- ]]does user activity on the system slow down serving as much
- ]]as when unaccelerated?
- ]](i.e. due to budget restrictions, we have to let our users
- ]]log in to the server, which at present is a bad idea...)
-
- >The load on the server is fairly minimal as its the card which does most
- >of the work.
-
- I have a Sparc II with PrestoServe, and running the eNFS software.
- I have noticed that it is now possible for even a single poky machine to push
- the load on the server up to 2! Even our dinky Sparc I's and the HP 9000/835
- after upping the number of biods to 6 did this. Obviously the client is
- getting quick service since the enfsd's on the host aren't sitting blocked
- most of the time. Everything zips into memory, and thus on the server
- you actually see enfsd's taking up real CPU time (like 10% each) if you
- watch top. An HP 9000/750 with 6 biod's can push the server up to 4 in
- my experience.
-
- Obviously this may have an impact on users logged into the server :-)
-
- --
- The War on Drugs is just part of the War on the US Constitution.
-