From: Colin Barber (Colin.Barber@telewest.co.uk)
Date: Wed Mar 05 2003 - 18:55:25 GMT-3
The router will almost always be buffering to a small degree. If you have a
fast Ethernet interface then it's quite easy to send a burst of data that
will fill the serial interface buffer, even if the line was at 0%. If you
increase the buffer size then packets should not get dropped but latency
will increase, either way users will see a performance drop. This normally
is not too much of a problem as the TCP windowing should not allow too much
data to be sent at once without receiving acknowledgements. You will get
more problems if using connectionless protocols (UDP or IPX).
When the line has a high average load then the buffers will be heavily used
trying to bridge the miss-match of transmission speeds between the LAN and
WAN. Any new packets arriving not only have the latency of the line but also
of the latency of getting to the front of the buffer. This is why WFQ is
default on slow speed links so that small packets (which normally contain
user interactive traffic) are serviced quickly.
If the link only has one flow of traffic, e.g. a large FTP transfer, then
your thinking is correct. The remote end will be getting packets at a
constant rate and the higher the load, the more efficient the transfer. But
when there are multiple flows a burst from one user fills the buffers and
then other users packets have added latency.
The maximum throughput of a router is not always bus speed it's normally
processing power. If the throughput is causing the CPU to run flat out then
you will get drops on all the interfaces and not just the ones that are at
high load.
Colin
-----Original Message-----
From: Sam Munzani [mailto:sam@munzani.com]
Sent: 05 March 2003 19:17
To: Colin Barber; ccielab@groupstudy.com
Subject: Re: At what percentage of T1 utilization you should see performan
ce degrade?
My understanding was a full T1 is 1540000 bps. This is full duplex. That
means up to that speed if router keeps putting bits to the serial
interface's output bucket, the line will pick it up and forward it to the
other end. When the router starts buffering data more than that rate, the
serial interface buffers will be full and hence start dropping bits to the
bit bucket. I didn't see any issue until the bucket gets full. The drops are
added to the "output queue drop" counters on "sh int" command. My logic is,
when the T1 is at 80% utilization, the line can still forward every single
bit of it so nothing will go to the queue drop. Is my thinking wrong?
It's different issue that router's bus can handle only 75000 pps. So if my
3640 router has 2 Fast E ports and the traffic between those 2 interfaces is
high, it will consume router's bus speed to 75000pps and hence the queue
drop will happen on Fast Ethernet's input queues. In this case the router's
back plane is the bottleneck.
In short, if you are not hitting back plane limitation, your users should
not see any performance degradation either T1 is at 80% or at 100%. Because
every single packet/bits is getting forwarded properly and there is no need
for retransmission because of collision or CRC.
Am I thinking right or wrong?
Thanks,
Sam Munzani
CCIE # 6479(R&S, Security)
> 75,000 packets per second with each packet 64 Bytes is a throughput of
> around 38Mbps. How many of your packets are going to be just 64 Bytes? If
> packets are larger then the throughput goes up.
>
> What does happen though on a slow serial link that is utilised above 80%
is
> there is not enough bandwidth to cope with bursts of data without
buffering,
> and maybe dropping, packets. This increases latency and also could causing
> problems with routing updates/hello packets being dropped. This is why
Cisco
> recommend a maximum average of 80% utilisation on a serial link.
>
>
> Colin
>
> -----Original Message-----
> From: Sam Munzani [mailto:sam@munzani.com]
> Sent: 05 March 2003 15:51
> To: ccielab@groupstudy.com
> Subject: At what percentage of T1 utilization you should see performance
> degrade?
>
> Hi,
>
> Today we got in to a debate about T1 saturation and performance
degradation.
> Below is 2 different opinions from my self and other engineer.
>
> What I say.
> T1 is a full duplex mechanism. So until it hits 100% utilization on it's
> Serial(Full T1) interface it will forward to packet and no packet drop
will
> happen. Only the time it will do a queue drop is when the traffic bursts
> above
> 100%. This queue drop is because of Serial interface short on forwarding
> buffers. Same behavior will happen on inbound. Only the time this would be
a
> bottleneck is if router CPU is a bottleneck. This is on 3640 with Fast
> switching turned on. I am assuming 3640 can handle more than 1 T1
bandwidth.
>
> The other engineer's opinion.
> 3640 supports only 75000 pps with 64 bytes packets. His opinion is, your
> users
> will see performance problems even before T1 is hitting 100%(aroung 70 to
> 80%).
>
> Any opinions with enough supporting proofs are most welcome.
>
> Regards,
>
> Sam Munzani
> CCIE # 6479(R&S, Security)
>
> --------------------------------------------------------------------------
---- > Live Life in Broadband > www.telewest.co.uk > > > The information transmitted is intended only for the person or entity to which it is addressed and may contain confidential and/or privileged material. > Statements and opinions expressed in this e-mail may not represent those of the company. Any review, retransmission, dissemination or other use of, or taking of any action in reliance upon, this information by persons or entities other than the intended recipient is prohibited. If you received this in error, please contact the sender immediately and delete the material from any computer. > > > ============================================================================ == >------------------------------------------------------------------------------ Live Life in Broadband www.telewest.co.uk
The information transmitted is intended only for the person or entity to which it is addressed and may contain confidential and/or privileged material. Statements and opinions expressed in this e-mail may not represent those of the company. Any review, retransmission, dissemination or other use of, or taking of any action in reliance upon, this information by persons or entities other than the intended recipient is prohibited. If you received this in error, please contact the sender immediately and delete the material from any computer.
==============================================================================
This archive was generated by hypermail 2.1.4 : Sat Apr 05 2003 - 08:51:33 GMT-3