| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1191.1 | is disk throughput a concern? | MUDDY::WATERS | | Thu Dec 30 1993 10:25 | 14 |
| >DEC 3000/800's tied together via FDDI and Gigaswitch. They have a requirement
>to move some large data files between systems with a time constraint. They
>are looking at using TCP/IP as a protocol. I would like to find some
>performance data on throughput over FDDI using TCP/IP to determine
Though I don't have the details, UDP on DEC 3000/500 saturates an FDDI
(over 95 Mb/s). For DEC 3000/800, I expect that TCP/IP will saturate
one FDDI link (> 95 Mb/s), and possibly go much higher (fill 1.7 FDDI
links, for example).
I have not heard of anyone with a disk configuration that sustains
95 Mb/s, e.g., using NFS and UDP. You said "data file". If you really
mean "file", and not "data block in memory", worry about the disk
subsystem before the network stack.
|
| 1191.2 | | QUIVER::STEFANI | Have the # for the Mars Observer? | Thu Dec 30 1993 14:19 | 6 |
| There is a whitepaper on the subject of using TCP/IP over FDDI on Alpha
systems. It's on the gatekeeper.dec.com FTP server:
/pub/DEC/DECinfo/whitepaper/hiperf-tcp.ps
- Larry
|
| 1191.3 | Real Issue is Network Contention | DVOPAS::SQRONE::HOOVER | Greg Hoover - DMD Sales Support | Mon Jan 03 1994 11:57 | 4 |
| The real issue that the customer is concerned with is effective thruput with
multiple systems using the FDDI network. The paper mention in .-1 is nice for
a point-to-point maximum thruput, but we are really trying to find out
what kind of thruput we can expect on a loaded network.
|
| 1191.4 | | KONING::KONING | Paul Koning, B-16504 | Mon Jan 03 1994 12:17 | 4 |
| That's a good reason for using a Gigaswitch: you get full dedicated FDDI
performance on each of its ports...
paul
|
| 1191.5 | Too many systems to rely on Gigaswitch!! | DVOPAS::SQRONE::HOOVER | Greg Hoover - DMD Sales Support | Mon Jan 03 1994 15:38 | 13 |
| >That's a good reason for using a Gigaswitch: you get full dedicated FDDI
>performance on each of its ports...
Yes, that would be nice, but in this solution there are more than 34 systems to
deal with. The Gigaswitch in this solution will serve multiple rings with
multiple systems on each ring. We really need to get end-to-end performance data
on a single ring under contention by the multiple systems.
It would be ideal if I could see some kind of performance characterization of the
end-to-end thruput on a ring based on number of simultaneous systems using the
bandwidth of the ring. I know I won't be able to see exact numbers to match this
particular configuration, but it would be helpful to see what happens to the
95Mbps when there are 10, 20, 30, ... simultaneous "file transfers" occurring.
|
| 1191.6 | | KONING::KONING | Paul Koning, B-16504 | Mon Jan 03 1994 16:47 | 6 |
| The total capacity should be unchanged independent of the number of active
clients. If they are all comparable, they should all get an equal share.
If some have small queues and others have big queues, the ones with the
big queues will tend to get a bigger share.
paul
|