Has it become common usage to define latency in an IP network as the round
trip time in that network?
I've always considered latency to be a one-way measure of delay and RTT to
be the sum of the latencies in both directions. When I tried to find
something to back up this view, I discovered that a number of companies
define latency as equivalent to RTT in their SLAs.
Assuming that one has measuring devices in every PoP, do you think it is
harder to measure a full matrix of one way latency compared to measuring a
full matrix of RTT?
Does it even make sense to measure a full matrix of RTT when the
measurement of A to B to A should be equivalent to the measurement of B to
A to B?
--Michael Dillon
Ah but its not is it, so much of the Internet is asymmetric now this is
important especially for troubleshooting
But generally performance relates to how well say your web pages download from a
site which is a 2 way tcp connection to you may as well do rtt and get both
directions.
Your problem with 1-way is finding a way to do it thats reliable, for rtt ping
is about as simple as you can get and works a treat but for 1-way you have to be
a bit more creative....
Steve
Could this be because customers believe that RTT more closely
represents a network characteristic that they (or their customers)
actually need? Perhaps they find it easier to digest like this
because they're already familiar with the concept (because someone
let them have 'ping').
Has it become common usage to define latency in an IP network as the
round trip time in that network?
It probably depends on the context.
Assuming that one has measuring devices in every PoP, do you think it
is harder to measure a full matrix of one way latency compared to
measuring a full matrix of RTT?
Yes probably, because of the potential assymetric paths, changing paths
and synchronization issues. See below.
Does it even make sense to measure a full matrix of RTT when the
measurement of A to B to A should be equivalent to the measurement of
B to A to B?
I think so. It would probably be nice to do both if you could though.
RTT may be particularly nice to measure, because the path between A to B
or between B to A may not be stable. In addition, the endpoints may not
be stable either (e.g. if some type of load balancing is happening).
The problem with measuring one way latency is that the receiver has to
be synchronized with the sender. The receiver also has to either
perform the collection of the data or return at minimum the received
timestamp. This is all a little more complicated than a simple 'echo'
or 'ack' service used for RTT measurements.
John
Has it become common usage to define latency in an IP network as the round
trip time in that network?
I've always considered latency to be a one-way measure of delay and RTT to
be the sum of the latencies in both directions. When I tried to find
something to back up this view, I discovered that a number of companies
define latency as equivalent to RTT in their SLAs.
Assuming that one has measuring devices in every PoP, do you think it is
harder to measure a full matrix of one way latency compared to measuring a
full matrix of RTT?
The problem is buying and installing the equipment, even if you buy an off
the shelf product like RIPE NCC's TTM :-). Once installed, these products
will just provide you with the numbers.
Does it even make sense to measure a full matrix of RTT when the
measurement of A to B to A should be equivalent to the measurement of B
to A to B?
If you are sure that the path taken for A-B-A is equal to B-A-B, then no,
measuring only A-B-A is sufficient.
Henk
Perhaps folks would be interested in the products of the IPPM WG
of the IETF. Although there's a lot of theory to wade through,
there are also open source tools referenced in the email archive.
No need to re-invent.