Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 4 Oct 2007 12:02:56 -0700
From:      Michael DeMan <michael@staff.openaccess.org>
To:        freebsd-net@freebsd.org
Cc:        lists@codeangels.com, Cristian KLEIN <cristi@net.utcluj.ro>
Subject:   Re: FreeBSD as a gigabit router
Message-ID:  <0D18E826-52EA-4BEC-9404-1C98BFCDD418@staff.openaccess.org>
In-Reply-To: <470535D6.7020601@net.utcluj.ro>
References:  <4703F9C3.2060601@net.utcluj.ro> <4532.192.168.2.137.1191451931.squirrel@www.codeangels.com> <470535D6.7020601@net.utcluj.ro>

next in thread | previous in thread | raw e-mail | index | archive | help
Hi All,

I've done some ad-hoc testing off and on for a few years.  None of  
the data around, but we do have a couple rules of thumb that we use  
internally...

1) Get the fastest PCI bus you can - PCI-X, etc.

2) Plan on 1GHz of CPU per 1 gigabit of throughput.

The performance hit going from FBSD4.x to FBSD5.x/6.x was horrendous  
for this kind of stuff, hopefully 7.x will speed things up.

Also, thus far, we have stuck only with single CPU machines to be  
conservative/safe.  We are looking forward to a speedier TCP/IP stack  
in 7.x and hoping to go to SMP routers at that time also.

Also, we've noticed at least on FBSD 6.x that there seem to be very  
few advantages in using polling on network interfaces.  We still run  
it, so that we have responsive SSH/BGP/OSPF processes on the  
machines, but my testing has shown that for sheer throughput, there  
is basically no difference.  I'd be curious if anybody knows the  
scoop on this.

Thanks,

- mike

Michael F. DeMan
Director of Technology
OpenAccess Network Services
Bellingham, WA 98225
michael@staff.openaccess.org
360-733-9279


On Oct 4, 2007, at 11:49 AM, Cristian KLEIN wrote:

> Thank you all for your replies.
>
> Kirill Ponazdyr wrote:
>>> Hi list,
>>>
>>> A few days ago I tested whether a FreeBSD 7 box is able to handle  
>>> Gigabit
>>> Can anybody point me what the bottleneck of this configuration  
>>> is? CPU was
>>> mostly idle and PCIe 1x should carry way more. Or is the experiment
>>> perhaps
>>> fundamentally flawed?
>>
>> ICMP is not a good way to perform such tests as many have mentioned,
>> better use iperf.
>
> I used this test, because it proved perfect when, almost a decade  
> ago, gigabit
> appeared. There wasn't anything at that time that could fill 1  
> Gbps, so we used
> the routers themselves to do the job. Also, I used this setup to  
> avoid TCPs
> congestion control mecachnism and sub-maximum bandwidth.
>
> Of course, when I said "ping -f", I didn't mean a single "ping -f",  
> but rather
> enough ping -f so that the looping packets would saturate the link.
>
>> We have a FreeBSD 6.2 / pf box handling 2Gbps of traffic, real  
>> traffic, it
>> will probably handle more, we just had no capacities or need to test.
>>
>> Hardware is a Single 2.4 Ghz Xeon with 2 x Intel Quad Pro 1000MT  
>> PCI-X
>> Controllers on separate PCI-X Busses.
>
> Could you tell me, is there any difference between 1000PT and  
> 1000MT, except the
> slot type? Also, is there any difference between Intel Desktop and  
> Intel Server
> adaptors, or are these just marketing buzzwords?
>
>
> _______________________________________________
> freebsd-net@freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-net
> To unsubscribe, send any mail to "freebsd-net-unsubscribe@freebsd.org"
>




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?0D18E826-52EA-4BEC-9404-1C98BFCDD418>