Hello all, I've been looking at optimizing my performance (both throughput and CPU wise) for Samba file sharing. I have some rather interesting results... My tests were done with a 1.9GB file on the server and a 1.8GB on the client. Here's the numbers: Reading the 1.9GB file from the server, I could attain a peak of 7.1MB/sec throughput, but it was on average closer to 6MB/sec. Writing the 1.8MB to the server was another matter. At one time I was able to achive a 7.1MB/sec peak throughput as well, but the averages were quite a bit lower, e.g 4MB/sec. One interesting note here is that sometimes I would only get around 2MB/sec. So here's the oddity. In tweaking parameters and such, I was restarting smbd between smb.conf changes for good measure. My 4MB/sec write rate would dramatically drop to 2MB/sec and I assumed it was from my parameter change. Undoing my parameter change and restarting still left me with a 2GB/sec write rate. Strange... Oddly enough if I restart smbd I will occasionally get the 4MB/sec transfer rate, and other times get 2MB/sec. I can't figure out why it's different one time versus another. I've tried it with different client machines, same situation. I've even had one work at 4MB/sec and the other at 2MB/sec... It almost seems to me to be something about the servers TCP/IP connection, but I really can't see how. Specifically, I'm running RedHat 6.2 on a dual PPro 166/512K cache and I've tried this with the included Samba 2.0.6 and my own compile of 2.0.7. I've also used the Stock Redhat Linux-2.2.14-XX kernel and the v2.4.0-test9 kernel. Same effect on all variations. So, despite my oddity, I have to say I'm quite impressed. Transfering directly between my client machines running Win98 (Celeron 550 and PIII 450) I didn't get nearly as nice of a transfer rate as w/ the server. And the clients always seem to use 100% CPU while the Samba server only uses about 25-30% of the server CPU. Very nice. Ah, and my benchmark numbers were acquires using a network monitor from AnalogX, so they aren't quite file MB/sec, but raw TCP/IP MB/sec (I think.) Any ideas out there? I'm going to keep trying to figure it out, it just got too late last night for me to keep trying... BTW: why aren't there any good 100MB network benchmarks out there? All I seem to find is some 10MB network benchmarks which aren't bad, but I'd like to compare apples to apples... (x86's to x86's?) Cheers! -= Shane.Anderson@infraredsolutions.com ======= 763-551-0003x229 =- -= Sr. Electrical Engineer ==== http://www.infraredsolutions.com =-