Discussion list for thrulay, a network testing tool

["Logg, Connie A." <>]

Sure...you can forward my message.

There are several options we are going to try...set the MTU to 9000 for 
example.

I will try your suggestion and get back to you.

We are not too sure if the 10gig nics/drivers are working 
correctly/optimally. We cannot get iperf and bbdp to work on these machines.  
But I did want to let you know we got thrulay installed and have made progess.

Connie
Connie Logg, Network Analyst
Stanford Linear Accelerator Center
ph: 650-926-2879 
"Happiness is found along the way, not at the end of the road, and 'IF' is 
the middle word in life."


-----Original Message-----
From: stanislav shalunov 
[mailto:]
 
Sent: Tuesday, October 18, 2005 12:18 PM
To: Logg, Connie A.
Cc: Cottrell, Les; McKee, Shawn; Li, Yee-Ting; 

Subject: Re: Thrulay @ sunnyvale

Connie,

Excellent.  Does this meet your performance expectations (1.36Gb/s 
single-stream and 2.31Gb/s with multiple streams)?  (I don't know what the 
hardware is capable of.)  How does it compare with other testing programs?

I suspect performance could be improved by setting the block size higher than 
the default 8kB, maybe -l131072 or even more---this would need 
experimentation.  Would you be willing to run several tests as follows?

        for l in 8192 16384 32768 65536 131072 262144 524288 1048576;
        do thrulay -l $l hostname | fgrep "**"; done

(This should take about 8 minutes to run and produce 8 lines of
output.)  The results of this 8-minute test would help us tune the default 
block size better, if needed.

An additional performance-enhancing avenue is TSC-based timekeeping, which 
would halve the number of context switches in thrulay operation.

                                                              --Stas

P.S. Is it OK to forward your message to 

 and copy this exchange there?  I think this might well be of general 
interest.