Hello!
I have some AT&T MPLS sites under a managed contract with latency
averaging 75-85 ms without any load. These sites are only 45 minutes
away. What is considered normal/acceptable?
Thanks,
I recently had a scenario with some MPLS sites within the state nearly
doubling in latency (below 50ms round trip). Typically I see
round trip latency in the 20-35ms range, and those sites are within about a
90 minute drive from each other (Oklahoma, mostly T1s).
When I asked an AT&T tech to investigate, he did not see log entries to
explain the increase or admit to any trouble, and stated that the service
levels for these MPLS circuits allowed for 75-80ms and I don't recall if
that was one way or round trip. He said that was to allow for coast to
coast latency scenarios. Delay returned to typical levels about 4 days
later, without explanation.
Acceptable from a technical standpoint (in that stuff works) or acceptable
from an expected service standpoint?
In the case of the former, MPLS can run over really high latencies, so
you're nowhere near the limit.
For the latter, 85ms would be highly unacceptable to me for a circuit to a
site that's so close. I would think your traffic is either being routed
really, really badly or their circuits are way over-subscribed.
MPLS as a technology should not add any significant delay as it is just a few bytes of label on the packet.
What is the physical path of the circuits involved? Have you asked for the design of them?
- Jared
Assuming no configuration errors, this underscores the need to negotiate
SLAs, and serious SLA penalties, with the telcos, and to always request a
telco network map, with the telco path that data will be transitting
end-to-end.. My rule of thumb in network design is that data over copper or
fiber takes 10 ms per 1000 miles, which is governed by the speed of light.
Network devices along the path add serialization/de-serialization delay,
but with modern network devices this delay is negligible. So according to
this rule of thumb 85 ms is almost enough time for data to traverse the USA
3 times.
I have found that telcos have been setting round trip SLAs so high that
they are meaningless (e.g. 50 ms for a GigE MEF ELAN service, 20 ms for
"Gold" MEF EVPL service), and border on being fraudulent. In one case I
also noted 100 ms round trip times between sites less than 1 mile away, and
discovered that every packet was being sent back to east Texas from
Southern California, almost a 5000 mile detour.
I have some AT&T MPLS sites under a managed contract with latency
averaging 75-85 ms without any load. These sites are only 45 minutes
away.
I've noticed this with AT&T's MPLS product when dealing with the
internal corporate network here. I don't know what they're doing wrong
but it is so very wrong.
What is considered normal/acceptable?
Less than 10ms unless you're using a sub-T1 interface or going a very
long distance.
Regards,
Bill Herrin
From: William Herrin <bill@herrin.us>
Subject: Re: MPLS acceptable latency?
To: "Mikeal Clark" <mikeal.clark@gmail.com>
Cc: "NANOG [nanog@nanog.org]" <nanog@nanog.org>
Date: Thursday, November 15, 2012, 1:23 PM
> I have some AT&T MPLS sites under a managed
contract with latency
> averaging 75-85 ms without any load. These sites
are only 45 minutes
> away.I've noticed this with AT&T's MPLS product when dealing
with the
internal corporate network here. I don't know what they're
doing wrong
but it is so very wrong.
circa 2007, noticed same thing: never below 90ms coast-to-coast across as13979. atm ds3 handoffs on both ends.
Perhaps the network is "oldish" and there are BW bottlenecks that lead to queues on the switches/routers that results in higher latency.
This would depend alot on the internal QoS strategy used by AT&T, the type of equipment used and the load in different parts of the network.
The only way to know what happens inside their MPLS cloud is to get past support and ask someone from the technical staff.