Quick Question:
We just installed a T3 from our 7204 to a major backbone provider and
I'm seeing 20 ms between us and them when I ping right from our router to
theirs. I also see it consistanly for that hop on traceroutes to and from
us. That seems to be rather a lot of latency for that line. The thing that
bothers me is that it's a consistant 20 ms when the line isn't loaded. Any
thoughts?
Chuck Scott
cscott@gaslightmedia.com
What distance is it running ?
In response to all the questions below...
The distance is from Northern Michigan to Chicago, apparently via
Detroit, which is about 500 Miles without knowing which way the fiber
really goes. There are I believe more than one SONET ring involved in the
transport. Also something like 13 cross connects, but no other routers in
that path.
When asked, the provider suggested that the latency was due to the
router I'm hitting being fairly busy. However, the latency is that same
when I ping it as when I go through it and it's always 20 ms, nothing
less. I'd think that if it's a matter of a low priority response issue,
that the latency would be variable and not be a fixed addition to the path
times going past the router.
OK, here's the funny thing, I have an account on anohter system here in
town that's also connected via a T3, but is 7 hops to the router that is
my first hop. When I ping the router that's my first hop from that system
I get about 22ms for all 7 hops while I get 20ms for just the first hop.
I've been looking for any system that would have low latency response on
the other side of that router, but so far nothing on the other side of it
is anything less than 20ms.
I guess the other question is how much of a marketing liability is this
going to be for my service. We're spending the money on this line to get
us the best connectivity we can from up here. Something tells me that some
dedicated or co-location customer is going to ask me about this latency
issue.
Still wondering...
Chuck Scott
--From Chris
What distance is it running ?
--From Jeff
You may want to check with the carrier of the circuit and make sure
that it's taking the path you expect. If it's on someting like a SONET
ring, it may be riding a much longer path that you would expect.
--From Brian
keep in mind you are pinging/tracerouteing that is aimed at the router.
ICMP is very low on the routers priority
list. The major providor's aggrigate router is prob pushing a few loaded
links. Better test is ping to a idle host nearby off their router.
--From Jonathan
With certain routers if they have enough stress, the RTT to the router
will be larger then though the router. This has to do with caching
algorithms the router uses. Also, some large providers do not have you
connect directly into a rotuer, but into a switch that acts as a MUX, and
then they have an OC48 link up to the router. If that OC48 has been
oversubscribed you might see latency, though I would hope not that
consistent.
What does your ISP's install engineer say? For a T3 that is going less
then 100 miles, the latency really should be 10percent of what you are
seeing.
Charles,
One thing I have a hard time explaining to some customers is that
latency is one thing.... what does it tell me... it tells me that from
one hop to another things are a bit slow.... the real important thing is
how are you're throughput speeds... I started a thread a while back asking
a similar question... is ping/traceroute a good measurement of throughput
on the link? the unanimous response was use pathchar or mtr or ttcp which
all give you a better guestimate of how your link is doing performance
wise..
Thanks,
Paul
///////////////////////////////////////////////////////////////////////////////////////////////////////////
A little nonsense now and then is relished by the wisest men.
-- Willie Wonka
\
Paul A. Bradford CCNA Adelphia High Speed Data
Voice: (814)274-6663 Network Engineer II
Fax: (814)274-0780 paul@adelphia.net
ICQ #6054021 pbradford@adelphia.net
That is an important thing by itself. While throughput is nice too,
latency, and to the same extent jitter control, are important to many
real time application users (e.g. gamers are a big category right now).
David Reed put it nicely when he said:
http://www.contextmag.com/archives/199907/TheGreatLie.asp
John
Sally,
I've starting monitoring packet drops on a couple of my routers. In
particular, our Internet router, which currently has 18 Mb/s on the
WAN side and 100 Mb/s side, I'm going to put RED on (the WAN side) for
outbound traffic.
I'm hoping RED can help keep the output queues lower and more stable
than they are now in order to improve overall latency/jitter between
us and our ISP.
I was going to use the parameter suggestions you have outlined in
REDparameters.txt. I'm currently using a Cisco 7206 running IOS
version 12.0(11.6)S on our border router. I was planning on
setting the parameters the same for all classes of traffic.
Our Internet WAN link is currently being monitored here:
http://netmon.depaul.edu/mrtg/uunet/rtrinternet.depaul.edu.6.html
and I recently set the following page to watch packet drops:
http://netmon.depaul.edu/mrtg/aqm/rtrinternet.depaul.edu.1.html
Note, this machine's clock is one hour ahead. I need to talk to
the sysadmin for this system to get that fixed one of these days.
Do you have any thoughts on my RED testing? Do you want to be
updated on the progress over the next few weeks?
FYI... we'll be moving our Internet connection to the Chicago NAP soon
and at that point I'll be able to test RED on a much higher speed
link (OC3c).
John