""Frank J. Manion"" manion at AQUILA.RM.FCCC.EDU
Fri Jun 16 08:34:49 EST 1995

>Does anyone use NFS for access to the databases? I ran some tests on NFS 
>access to databases and the results were horrendous. A DEC 3000/600 connecting
>to an AlphaStation 2/233 (A.S. acting as server) would "hammer" the network
>when 1 search was running. Increase the number of searches and the network 
>traffic decreased. I guess this is due to the fact that the NFS client isn't
>requesting data at the high rate. Interestingly, if more than one search was
>running on the NFS client the actual time taken for the search was identical
>to the time when the data was local. One search running with NFS mounted data
>was, on average, 10-20% slower than normal. Now we are considering moving to
>FDDI so does anyone have suggestions to the performance improvement? 

We have been reluctant to mount the databases with NFS for similar reasons.
We've done some quantification of network traffic in general between
AXP platforms (3000/300 and 3000/400's) using HP LanProbe network monitoring
equipment and have found up to 80% network saturation with these machines. Since
ethernet is a contention-based protocol, its efficiency starts to degrade
rapidly at the 50% mark. Our experience with FDDI has been limited to date to 
Sun and Ultrix boxes, but on these platforms we experienced a two-fold speed 
increase in large transfers.  Ultimately this sort of network traffic probably 
demands switched transfers of some sort, either eithernet (fast or normal) or 

I suspect that newer, faster processors from all manufactures are soon going to 
be a major problem for 10-megabit ethernet in general without careful network
planning and hierarchical switching/bridging.

-- Frank

Frank J. Manion
Director, Research Computing Services
The Fox Chase Cancer Center                              Phone: 215-728-3660
7701 Burholme Avenue                                     FAX:   215-728-2513
Philadelphia, PA  19111                         internet: FJ_Manion at fccc.edu

More information about the Info-gcg mailing list

Send comments to us at biosci-help [At] net.bio.net