[prev in list] [next in list] [prev in thread] [next in thread] 

List:       zebra
Subject:    [zebra 22486] Re: 40K pps
From:       Michael Bernstein <mb_jobs () yahoo ! com>
Date:       2005-06-30 14:14:37
Message-ID: 20050630141437.27624.qmail () web33410 ! mail ! mud ! yahoo ! com
[Download RAW message or body]

Hi I was wondering if you can help me understand.
There were a few things in this last message:

When you're talking about linux pktgen (you say
testing or can it be used in production?). How does
this fit in with Zebra?

Now, when you talk about a PCI-X slot in a
motherboard. For example, my ASUS board has one but
I've never used any type of PCI-X card before. I'm
guessing what you're talking about with "PCI-X bus
even 1.488 Mpps" means you're using a GigE or FastEth
PCI-X card and achieving extremely high packet
processing because the PCI-X bus is so much faster
when compared with regulard PCI or SCSI. The bus is
what allows the CPU to talk with the interface card,
so a faster and wider bus means more speed.. Am I
right here about any of this? Thanks for cluing me in.

The last questions were in regards to rDoS (what is
it?) and LC-trie (what is this?).

Also, when you mention "flow testing" what exactly do
you mean by this?

I really am interested in learning more about routing.
Most of my background is in Cisco.

Thanks again,

Mike



--- Robert Olsson <Robert.Olsson@data.slu.se> wrote:

> 
> Sean Reifschneider writes:
> 
>  > The Linux kernel includes a packet generation
> program that can be used for
>  > testing.  This is what I'm thinking about using
> that to see how many
>  > packets I can generate to push over a test
> router.  I have a 3GHz system
>  > coming in later this week that I can use for that
> testing.
>  
>  Yes pktgen can send very high rate for pps-testing
> with some patches
>  to e1000 on high speed PCI-X bus even 1.488 Mpps.
> ipv4 as ipv6 
>  and even do rDoS and flow testing.
>  
>  Also note all NAPI drivers in Linux turns into
> polling when load is high
>  the reason for this design is that we don't want to
> spend resources 
>  CPU and bus on devices that has no or little
> traffic. This means we can
>  build systems with many NIC's.
> 
>  Below an output from PIII 933 MHz linux router with
> full BGP (2 views) 
>  e1000 NAPI
> 
> ifstat2 eth*
>              RX --------------------------   TX
> -------------------------
> eth0          685.4 M bit/s       59 k pps     18.1
> M bit/s       34 k pps 
> eth1            5.1 k bit/s        7   pps      6.8
> k bit/s        3   pps 
> eth2           18.0 M bit/s       34 k pps    683.5
> M bit/s       59 k pps 
> 
> rt_cache_stat 
>  size   IN: hit     tot    mc no_rt bcast madst
> masrc  OUT: hit     tot     mc
>  4199     98649      18     0     0     0     0    
> 0         6       0      0
>  4228     97757      14     0     0     0     0    
> 0         5       0      0
>  4252     95136      12     0     0     0     0    
> 0         3       0      0
> 
> Almost 100 kpps in total This box si in front of
> huge ftp archive so there are 
> not that many flows I should add.
> 
> Also I could say that LC-trie is now becoming an
> alternative for FIB lookup in 
> Linux kernel for ipv4. 
> 
> 
> Cheers.
> 						--ro
> _______________________________________________
> Zebra mailing list
> Zebra@ml.zebra.org
> http://ml.zebra.org/mailman/listinfo/zebra
> 


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 
_______________________________________________
Zebra mailing list
Zebra@ml.zebra.org
http://ml.zebra.org/mailman/listinfo/zebra
[prev in list] [next in list] [prev in thread] [next in thread]