[nsp] [article]: Filters on routers: The price of performance

[Ejay Hire]

I received a response to my query about the configs, and here are my
observations.  
Disclaimer:  I haven't had a chance to test these, but my gut tells me
this want a fair fight based on the following.

1.  The static routes are pointed to Ethernet interfaces, not hosts...
So the router will arp for every packet.
2.  The box didn't have CEF enabled, which most cluefuls would enable if
trying to maximize throughput.
3.  No logging console was not in the config, and every deny in the ACL
was set to log.  This will generate an interrupt for every ACL match
going through the box.

-Ejay


"Ejay Hire" <ejay.hire at isdn.net> writes:

> What I don't get is how they got the really shabby numbers from the
Cisco.  I just performed a similar test of throughput with a couple of
2501's back to back with large ACL's applied and didn't get results that
bad  (measuring throughput and latency with 64k and 1524 byte packets,
originated from a Linux box with traffic generator support, to various
source and destination ip's across a two serial interfaces with external
csu-dsu's configured to emulate t1 circuits.  Each serial interface
configured for hdlc with a 50 line ACL in place.).  Did they have
Multilink enabled and route-caching disabled with d cef turned off?
I've forwarded a request to the website for information about the
testing methodology.
> 
> -Ejay
> 
> -----Original Message-----
> From: Voralt [mailto:peder at voralt.net] 
> Sent: Tuesday, July 15, 2003 11:27 AM
> To: cisco-nsp at puck.nether.net
> Subject: Re: [nsp] [article]: Filters on routers: The price of
performance
> 
> Just another ridiculous magazine "real world test".  I like how they
don't
> take price into account, by comparing a $14k Lucent to a $11k Tasman
to a
> $6200 Cisco to a $3600 ImageStream.  If that's the case, then I would
say,
> compare apples to apples, set a max price of say $14k and then compare
the
> highest end prodct from each vendor at that price.  The Cisco 3725 is
> $10,500 and the 3745 is $14k (by their numbers).
> 
> Also, why did they use two 10M connections instead of one 100M
connection.
> That seems a little odd to me.
> 
> 
> ----- Original Message -----
> From: "Ejay Hire" <ejay.hire at isdn.net>
> To: <cisco-nsp at puck.nether.net>
> Sent: Tuesday, July 15, 2003 10:51 AM
> Subject: RE: [nsp] [article]: Filters on routers: The price of
performance
> 
> 
> > Hi all.  Anyone else take exception to this?  My real-world and Lab
> experiences with the 2600 series don't come anywhere close the latency
and
> throughput issues they describe.
> 
> -Ejay
>