Monday, November 17, 2008

GigE Benchmark, Pt. 2

I added a third node to the GigE backbone. Benchmarks for Adama and Mulder copying from Scully via NFS. Test file is an ISO of slightly under 1G size. The goal of the test it to determine the effect of processor speed on GigE throughput.

[root@mulder ~]# echo -n "`date +%T.%N` - "; \
  cp /mnt/isos/RHEL4-dvd-docs.iso .; date +%T.%N

09:40:46.403907606 - 09:42:23.930183317
-------------------------------------
[root@adama ~]# echo -n "`date +%T.%N` - "; \
  cp /mnt/isos/RHEL4-dvd-docs.iso .; date +%T.%N

14:08:53.343940918 - 14:10:12.375121976
(Obviously, somebodies clock is out of sync.)

Given that Scully is our "control value", we get:
  mulder = 1:37.53
  adama  = 1:19.03
Mulder has a 1G CPU and Adama has a 2G CPU. A minor difference that could have been caused by several factors. Now the big test. Scully is a PII 300Mhz. Lets see the transfer rate between Adama and Mulder.
  from mulder to adama = 1:22.48
  from adama to mulder = 1:03.18
Bottom Line
A system's CPU has a effect its transfer rate. Paradoxically, the best speeds come from having the faster system on the server side, rather than the client side. Of course the problem with this is that, in this case, the high speed system is needed for the application.

No comments:

Post a Comment