Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 14 Nov 1996 11:07:53 -0600 (CST)
From:      Joe Greco <jgreco@brasil.moneng.mei.com>
To:        shovey@buffnet.net (Steve)
Cc:        jdd@vbc.net, richardc@csua.berkeley.edu, isp@freebsd.org, chad@gaianet.net
Subject:   Re: Decision in Router Purchase
Message-ID:  <199611141707.LAA25384@brasil.moneng.mei.com>
In-Reply-To: <Pine.BSI.3.95.961114090207.17009Q-100000@buffnet11.buffnet.net> from "Steve" at Nov 14, 96 09:03:26 am

next in thread | previous in thread | raw e-mail | index | archive | help
> On Thu, 14 Nov 1996, Jim Dixon wrote:
> 
> > On Thu, 14 Nov 1996, Veggy Vinny wrote:
> > 
> > > > Don't buy a 2501.  You can get a lot more performance for the same 
> > > > amount of money using FreeBSD and sync serial cards.
> 
> I dont know why people are bashing the 2501's - with an access list
> active, my router stats on one of my T-1s has at times shown thruput very
> close the to max a T-1 one can handle going full guns.

Excellent!

Now try it with data packets that are somewhat smaller than 500 bytes.

My 386DX/40 with one of the ET cards was able to max out a T1 (at least in
one direction, the inbound channel was probably not maxxed), as long as the
data packets were moderate to large.  With 60% saturation my CPU was more
than 50% idle, but that number dropped rapidly as it approached saturation.

Once it started seeing more small packet traffic, it started to display the
dreaded slow-blinking-cursor syndrome that indicates you are way over what
the CPU is able to handle.

Now it is a 486DX5/133 with a pair of PCI DEC 21041's instead of a cruddy
NE2000, and I do not worry too much about it :-)

What is the CPU in a 2501, anyways?  I think a 68000 but that is not first
hand experience.  I have some Sun 3/60's with 68000's in them, anyone want
one?  They are roughly comparable to a 386DX/25.

On a very much related-to-routers note...

I do have a sad report.  Trantor, my mega-386DX/40-from-hell with 6 ISA
Ethernet cards, was taken out by a faulty hub (don't know exactly what
happened yet).  It was replaced (at least temporarily) by a nice ASUS
P55T2P4 Pentium 100 with a Znyx 314 and (alas) an NE2000...  

The following tests are purely routing, with several other boxes providing
the beating.

Doing a wire to wire FTP transfer with all 21x4x based cards gets me 
about 2500 packets per second, with a massive number of collisions.

 procs   memory     page                    disks   faults      cpu
 r b w   avm   fre  flt  re  pi  po  fr  sr f0 w0   in   sy  cs us sy id
 0 0 0 31712  3288    1   0   0   0   0   0  0  9 5587   38   8  0 26 74
 0 0 0 31712  3288    1   0   0   0   0   0  0  0 5013   34   7  0 25 75
 0 0 0 31712  3288    1   0   0   0   0   0  0  0 5106   30   6  0 26 74
 0 0 0 31712  3288    1   0   0   0   0   0  0  0 5304   30   6  0 28 72
 0 0 0 31712  3288    1   0   0   0   0   0  0  0 5435   34   7  0 33 67

I like the % idle :-)

Doing the same thing out the NE2000 based port gives the same throughput
but only 5% idle :-(  (Moral of the story: NE2000 sucks, but we all knew
that).

Now I try nailing it with "ping -f"...

 procs   memory     page                    disks   faults      cpu
 r b w   avm   fre  flt  re  pi  po  fr  sr f0 w0   in   sy  cs us sy id
 0 0 0 31612  3284    1   0   0   0   0   0  0  0 8390   32   6  0 36 64
 0 0 0 31612  3284    1   0   0   0   0   0  0  0 8392   42   6  1 40 60
 0 0 0 31612  3284    1   0   0   0   0   0  0  5 8068   32   7  0 35 65
 0 0 0 27352  3284    1   0   0   0   0   0  0  0 7045   34   7  0 36 64
 0 0 0 27352  3284    1   0   0   0   0   0  0  0 8392   30   6  0 36 64
 0 0 0 27352  3284    1   0   0   0   0   0  0  0 8029   30   6  0 41 59
 0 0 0 27352  3284    1   0   0   0   0   0  0  0 7533   30   6  0 33 67
 0 0 0 27352  3284    1   0   0   0   0   0  0  0 8657   30   6  0 39 61
 0 0 0 27352  3284    2   0   0   0   0   0  0  0 8869   42   8  1 37 62
 0 0 0 31528  3284    1   0   0   0   0   0  0  0 8789   42   8  0 38 62

We are handling about 4K packets per second and maybe now are just 
starting to break a sweat..  response is still _instantaneous_.

I tried a few other things and the numbers I got were varied.  I was not
doing a particularly scientific test, I just wanted to make it fall over,
and I couldn't.  The box was also handling my normal network traffic while
taking this pounding...

This has nothing to do with the T1 thing, I realize, but when we start
talking about the ability of routers to route.

... Joe

-------------------------------------------------------------------------------
Joe Greco - Systems Administrator			      jgreco@ns.sol.net
Solaria Public Access UNIX - Milwaukee, WI			   414/546-7968



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199611141707.LAA25384>