Re: (ASCEND) Poor TNT performance?

[Richard Silver <rsilver@deltacom.net>]

I have seen some testing from the Tolley group - A TNT's throughput
drops off significantly as the load increases. This is probably what you
are seeing happening. The TNT seems to suffer from a severe lack of
processing power. Adequate for a moderately loaded box, but not
otherwise. 

Simon Rainey wrote:
> 
> Tim,
> 
> >> We typically see around one call coming up or going down per second on
> >> each primary rate.
> >...
> >> Every site demands their full bandwidth and our observations are that
> >> most of them do indeed saturate their links almost continously during
> >> the day.
> >
> >Since there are 32 channels in an E1 circuit, this would mean that each
> >connection is lasting only about 30 seconds on average.  This is at odds
> >with the idea that the links are continously saturated.  If they are
> >saturated, they should not be going up and down twice a minute.
> 
> We set the idle timer on the remote routers to 30 seconds because our users
> have a capped charging scheme for the use of their ISDN lines. They will
> never pay more than a set figure no matter how many calls they make or how
> long they are connected for. In an attempt to reduce the number of backend
> lines we have to buy we set the idle time to be something low, and 30
> seconds seemed like a reasonable figure.
> 
> We have observed two class of users. The majority hammer their connections
> for most of the day with web access, but even then there are usage patterns
> which cause the bandwidth to drop to zero for 30 seconds or more. Then
> there are users who do little else but pick up their mail, and these
> account for most of the call transitions.
> 
> Analysing the RADIUS accounting logs we see a small but significant number
> of users that come and go very rapidly. Some have call durations as low as
> 5 seconds and their line is going up and down all day. We're looking into
> why this should be, but it means that averaged over all the PRIs we do see
> about one call come up or go down per second per PRI.
> 
> >> am I right in assuming that the TNT's IP routing is all done by the
> >> main CPU rather than by a hardware routing engine?
> >
> >No, the routing table is maintained by the shelf router and distributed
> >to each slot card, where the IP routing is done.
> 
> That's reassuring.
> 
> >> Performing a throughput test revealed that we could only get 20kbps over
> >> a 64kbps connection.
> >
> >What type of traffic (ICMP, TCP, UDP) with what size packets (64, 1500)?
> 
> We downloaded a 10MB file from one of our Unix FTP servers using the Win95
> FTP client. The Win95 host was connected using a UTP crossover cable to an
> Ascend P50. When the P50 was dialling into the loaded TNT we could not
> achieve more than 20kbps. Repeating the test with the P50 dialling into a
> less loaded TNT or a Cisco AS5300 we achieved 60kbps. The tests were
> repeated multiple times. At one point we were seeing less than 2kbps
> through the loaded TNT. Once the load dropped off (early evening) we
> achieved 60kbps consistently.
> 
> Regards,
> Simon.
> ++ Ascend Users Mailing List ++
> To unsubscribe: send unsubscribe to ascend-users-request@bungi.com
> To get FAQ'd:   <http://www.nealis.net/ascend/faq>

-- 
+--------------------------------------------+
| Richard Silver			     |
| Network Administrator			     |
| ITC^DeltaCom Internet & Data Group	     |
| Auburn AL				     |
| 800-239-3000 x 3506			     |
|--------------------------------------------|
| Lord, Please save me from your followers.  |
+--------------------------------------------+
++ Ascend Users Mailing List ++
To unsubscribe:	send unsubscribe to ascend-users-request@bungi.com
To get FAQ'd:	<http://www.nealis.net/ascend/faq>