Network traffic simulator

Saku Ytti saku at ytti.fi
Wed May 25 07:08:32 UTC 2016


On 24 May 2016 at 23:04, Spencer Ryan <sryan at arbor.net> wrote:

Hey,

> We are heavily invested in Ixia, they are very expensive, but if you need
> the kind of precision they provide they work very well.

I've used all big three, Agilent (back in packets and protocols), IXIA
and Spirent. And several smaller/cheaper like Anritsu, EXFO, etc.

I think Agilent was the best product for SP's. But you no longer can
purchase it. I was recently responsible for evaluating and purchasing
product for one company and shortlisted both IXIA and Spirent. Neither
one really was fully satisfactory after Agilent experience, but we
needed a solution so we purchased Spirent.
For that company QoS testing was one of the key use-cases and you
really can't do that in IXIA at all, as you cannot make it burst for
any significant time (i.e. not for even significant microseconds).
Testing QoS with precisely paced packets is not going to be useful.
Spirent can do this nicely, so we chose Spirent.
I'm not being nasty or funny, but I think best thing that ever
happened to Spirent was IXIA buying Agilent.

Both Spirent and IXIA has much poorer graphing capabilities than
Agilent had. In IXIA I mostly rely on exporting data and graphing with
GNUPlot instead. For example you cannot graph packet loss as
percentage in the tool itself. Which is huge annoyance to me, after
coming from Agilent. Many times in QoS testing you'd have EF, AF, BE
traffic, and you have expectation how many percentage in given
situation should given class drop, doing this in Agilent is a chore.

Agilent probably has best in the breed network with emulation
capabilities. And focus generally seems to be in protocol
testing/development where network emulation is tremendously useful.

As the platforms are very expensive, not many SPs are using them, so
they're not getting input from SPs what the boxes should be doing.
This market is very poorly tapped, there is large demand in the market
for proper testing equipment but it's just priced out of reach. I
believe Spirent and Agilent should sell the hardware at-cost, then
sell timed licenses, where maybe 1000h license would be today's full
cost. Large segment of this market might not use box at all in some
year and would generally only require modest hours from it.
Bit harder to justify the cost with low use, compared to vendors who
run them automated 24/7.


-- 
  ++ytti


More information about the NANOG mailing list