Discussion list for thrulay, a network testing tool

["Cottrell, Les" <>]

 Yee can you try this please. Thanks.

-----Original Message-----
From: stanislav shalunov 
[mailto:]
 
Sent: Tuesday, October 18, 2005 2:37 PM
To: Logg, Connie A.
Cc: Li, Yee-Ting; Cottrell, Les; 

Subject: Re: Thrulay tests on sunnvale 10 g nic machines

Connie,

The 4.5Gb/s result from a single TCP stream is pretty good and might well be 
the hardware limit, for all we know, but it might make sense to try to tune 
it even further.

This seems to suggest that, in your environment, which is about as high-end 
as it currently gets, there's no good reason to increase the block size 
beyond 64kB, and certainly not beyond 128kB.

We'll consider increasing the default block size (currently at 8kB) in future 
versions.  However, it'll make thrulay not work as well in low-speed 
environments (e.g., DSL or cable).  I'm not sure how to approach it.  Let the 
user tune?  Maybe let the user enter the specific environment?  (E.g., it 
could be an option to specify ``home
use.'')

In the meantime, and regardless of what happens with the default in the 
future, you should be able to improve the performance by specifying the 
-l65536 option for thrulay.

Would you like to try getting more performance out of it?  This part needs 
better integration, which we'll get to, but for now the next step would be to 
install tsci2-0.0.8 from http://sourceforge.net/projects/tsc-xluo

        ./configure
        make
        make install
        tsci2d        # start the daemon manually to see if it's working

It makes sense to run tsci2demo then.  Good output might look like
this:

method availability test:
        using method TSCI2_DAEMON
accuracy test:
        gettimeofday(): 1129671091.347090.
        tsci2_getimeofday(): 1129671091.347095.
speed test: using 100000 runs
        gettimeofday(): 451844 microseconds
        tsci2_gettimeofday(): 23889 microseconds 

NB: the method is DAEMON.  Good.

Then rebuild thrulay, starting with configure.  The output of thrulay's 
configure script should then include the following line:

checking for tsci2_init in -ltsci2... yes

If it says ``yes'' there, it means it'll get the time through the TSC-based 
library (which will be at least an order of magnitude faster).  If it says 
``no'' it means it's not finding the library and there will be no change.  
(This integration check needs to improve in the future, perhaps by going to 
binary packages with dependencies as the distribution mechanism.)

-- 
Stanislav Shalunov              http://www.internet2.edu/~shalunov/

But we must show them that they cannot terrorize the greatest nation on
the face of the Earth.  And we won't.       -- George W. Bush, 20011017





"Logg, Connie A." 
<>
 writes:

> Snv-slac1 -> snv-slac2 - new drivers on snv-slac2 and old drivers on 
> snv-slac1
> [cal@snv-slac1
>  ~]$ more ./thrulay.test.sh
>         for l in 8192 16384 32768 65536 131072 262144 524288 1048576;
>         do thrulay -l $l snv-slac2-10g.ultralight.org  | fgrep "**"; 
> done
> 
> #(ID) begin, s   end, s  Mb/s    RTT, ms: min  avg     max
> #(**)    0.000   60.001  534.071    0.335  214.230 11714.837
> #(**)    0.000   60.001  265.902    0.543  549.848 8870.102
> #(**)    0.000   60.001   62.638    0.795 2504.723 4952.227
> #(**)    0.000   60.000  448.885    1.265  265.358 5343.916
> #(**)    0.000   60.000  125.305    9.760 1194.817 4084.552
> #(**)    0.000   60.001  330.718    8.732  251.543 1010.475
> #(**)    0.000   60.001  442.841   15.696  225.664  755.208
> #(**)    0.000   60.000  683.947   21.321  132.727  170.333
> 
> Old drivers on snv-slac1 and new drivers on snv-slac2
> Snv-slac2 -> snv-slac1
> [cal@snv-slac2
>  ~]$ more ./thrulay.test.sh
>         for l in 8192 16384 32768 65536 131072 262144 524288 1048576;
>         do thrulay -l $l snv-slac1-10g.ultralight.org  | fgrep "**"; 
> done
> 
> 
> [cal@snv-slac2
>  ~]$ ./thrulay.test.sh
> #(**)    0.000   60.000 2210.400    0.029    0.718    1.794
> #(**)    0.000   60.000 3621.265    1.617    2.134    2.629
> #(**)    0.000   60.000 3877.231    0.061    0.391    1.233
> #(**)    0.000   60.000 4162.874    0.288    0.628    1.441
> #(**)    0.000   60.000 4549.251    0.290    1.673    2.156
> #(**)    0.000   60.000 4555.904    1.558    5.207   10.809
> #(**)    0.000   60.001 4618.377    2.348    8.120   12.997
> #(**)    0.000   60.002 3417.134   10.980 1438.668 1648.186
> 
> I rebooted snv-slac1 which installed the new drivers/kernel And the 
> results from snv-slac2-> snv-slac1 are greatly improved!!!
> [cal@snv-slac2
>  ~]$ ./thrulay.test.sh
> #(**)    0.000   60.000 2205.789    0.031    1.255    1.436
> #(**)    0.000   60.000 3543.700    0.100    1.130    1.774
> #(**)    0.000   60.000 3869.199    0.180    1.346    1.578
> #(**)    0.000   60.000 4306.131    2.159    2.677    3.425
> #(**)    0.000   60.000 4434.800    0.349    1.599    1.801
> #(**)    0.000   60.000 4731.790    0.964    1.328    2.169
> #(**)    0.000   60.001 4594.183    1.637    1.824    3.632
> #(**)    0.000   60.002 4055.930   10.758   14.473   16.480
> 
> Good news....