Some interesting plots

[Randall Stewart]

Pyun:

Nope, the test is between em1 <----> em0

The only way the msk0 would be used is in the
event of loss..via a t3 timeout... and there
are no timeouts that happen in this plot.

R

Pyun YongHyeon wrote:
> On Fri, Sep 29, 2006 at 01:29:37PM -0400, Randall Stewart wrote:
>  > All:
>  > 
>  > As you all may know I have been working on getting SCTP
>  > into Current..
>  > 
>  > I have been, of late, trying to tweak things to get
>  > the BEST performance out of the implementation...
>  > 
>  > I have moved my testing off between two 7.0 machines.. same
>  > code base on each updated Sep 25.
>  > 
>  > One is a 2.8Gig Dell SC1600 Xeon.. (hyper threaded CPU).
>  > The other is a P4D (2.8Gig .. slightly faster, true dual
>  >                      processor machine).
>  > 
>  > They are connected by two intel EM server cards like so:
>  > 
>  > 
>  > +----+                                 +----+
>  >   1  | em1 <---------------------> em0 | 2
>  >      | em0 <-----locallan--------> msk0|
>  >      | dc0 <-Direct Inet               |
>  > +----+                                 +-----+
>  > 
>  > 
>  > em1 has 10.1.2.12                   em0 10.1.2.21
>  > em0 has 10.1.1.12                  msk0 10.1.1.21
>  > 
> 
> [...]
> 
>  > One other note, I see TCP is only getting 250Meg or so on the
>  > same test (It can run either).. now it used to get close to
>  > the full pipe (Gigbit).. so is there some issue with the new code
>  > that was recently submitted?
>  > 
> 
> I'm not sure but it seems that you've used experimental msk(4) on
> CURRENT. ATM msk(4) has Rx performance issue. So if you get very
> poor receive performance it would be msk(4) issue.
> 


-- 
Randall Stewart
NSSTG - Cisco Systems Inc.
803-345-0369 <or> 815-342-5222 (cell)