On Fri, 7 Apr 2006, Keith A. Glass wrote:

>> scalable. scaleable to what? are you talking an Internet connection where
>> you have a need for multiple T-1 lines? multiple DS-3 lines? multiple
>> OC-12 lines? or are you talking local networks where you have 100Mb
>> ethernet? or gig ethernet? or 10gig ethernet? are you talking just a
>> couple of these networks or are you talking about dozens of these
>> networks?

> We have initial estimates of 300-500 GB/day in SMTP traffic alone, due to an
> application that typically sends data in via SMTP in 2MB bundles. But they
> ALSO want to up the resolution of the graphics inside the bundles, so we've
> been told to expect an order of magnitude jump about the time we start
> implementing in the 2008-2009 timeframe. And the data will tend to peak and
> valley a lot. . . So, realistically, we're talking an initial traffic of 3-5
> TB/day in SMTP alone.

5TB/day is a sustained 60MB/sec (1 1/2 DS-3's or so), given that you have
a lot of peaks it's reasonable to say that your peak traffic is 2-3x that
value. you are still talking about ~200Mb/sec of traffic.

this is comfortably handled with a P-III intel platform (a Nokia 740
appliance is this amount of power)

Sun has a checkpoint appliance that is Opteron based (defaults to 1.4GHz
processors, you can upgrade it) for about $30K. this is a very moderate
box by today's standard, but would handle the type of bandwidth
requirements you are talking about trivially

> We have multiple OC's coming in, bandwidth isn't the immediate worry, it's
> throughput. . .

again I need to ask for definitions. the best overall throughput is
generally achieved by spreading the load evenly and running things at max
capacity all the time. bandwidth requirements better represent your peak
requirements, but I think what you are looking for is responsivness (or
low latency). Even with that you should keep in mind that Internet use
imposes a latency overhead (cross country is 100ms, a dial-up to the local
ISP add 300ms), so you shouldn't let people get worked up about small
latencies within your network or your firewalls. On modern hardware even
dumb, forking proxies can end up with low enough latency that when added
to a moderatly complex network don't add a measurable response to the
end-to-end response time of the system.

David Lang

There are two ways of constructing a software design. One way is to make it so simple that there are obviously no deficiencies. And the other way is to make it so complicated that there are no obvious deficiencies.
-- C.A.R. Hoare

firewall-wizards mailing list