looking for a review of traffic shapers
pete at partnercomm.com
Wed Nov 26 18:35:01 UTC 2003
(caveat: I am not in sales - I'm a very happy PacketShaper user.)
Traffic shaping comes in many shapes and forms. Many places who need to do
traffic shaping need to do it down to the application level, so those apps
that are of a mission-critical nature get through, while those that can
wait are made to wait.
Trouble with the 'diskless' units is a) they can't distinguish enough
different applications to really be useful, b) they don't have control
over the incoming traffic, and c) they can't provide you with a historical
trend analysis (or event analysis) to correlate with issues that may be
happening on the network. Most make use of queueing, which can only
provide some control for outbound traffic, and at worst can lead to
further retransmissions and thus further congestion.
The CMU study between the Allot and the Packeteer devices was not well
done, and I would encourage you to look further at each of these devices -
especially since a full year has passed since those reports, and that the
testing didn't even look at the rate-limiting features of them, which is
what would make all the difference.
Certainly, the shaping devices out there are not inexpensive, but a box
that can shape 10MBit of traffic, broken out to over 500 different
classes, a PacketShaper 2500 can be had for well under US$10000.
The ROI on these devices is proven, and depending on the scenario, it can
be months, not years before the device has paid for itself. In many
situations, it is the *only* solution to keeping control of
ever-increasing bandwidth demands in non-ISP (just move packets as fast as
I would be more than happy to discuss my experiences with these units.
On Wed, 26 Nov 2003, William Caban wrote:
> On Tue, 2003-11-25 at 13:38, andrew at vivalibre.com wrote:
> > Note: delurk.
> > Some of the commercial traffic shaping devices reviewed here are tens of
> > thousands of dollars. For a smaller ISP (i.e. less than a DS3 of
> > aggregate upstream bandwidth), that kind of expense doesn't make sense--
> > but the need to control bandwidth consumption is still an issue.
> > Is anyone on the NANOG list aware of a disk-less Linux solution? One might
> > imagine a Knoppix-like bootable CD image (perhaps CD-RW, so config files
> > could be updated) that would turn an inexpensive Linux box into an
> > effective traffic shaping device, using tools like CBQinit, MRTG/RRDTOOL,
> > and a Webmin-like admin interface. The closest thing to this I've seen is
> > ETINC's BWMGR, but that's a closed-source solution and is still somewhat
> > expensive.
> > -Andrew White
More information about the NANOG