On 28/11/16 02:26 PM, Jonathan Morton wrote:
On 28 Nov, 2016, at 20:37, David Collier-Brown <dave...@rogers.com> wrote:
I'm using latency as the time from the request to the first response
transfer time as the time from the first response to the last response, which
may be 0, and
sleep/think time as the time from the last response to the next request, for a given
stream of requests and responses, AKA "transactions"
This is a useful set of definitions, though I would quibble that “transfer
time” should also be measured from the request, not from the first response.
Latency + Transfer Time add up to make "Response Time", which is from
very beginning to very end, as you note.
A more complete picture would add a measure of application-induced delay, which
is visible to the user but not necessarily to the network. It is influenced
mainly by the end host's performance, not by anything in the network, and would
presumably count as part of the “think time” for many applications. Hence it’s
not relevant for *our* objectives, but might be to others.
The response also breaks down into queue delay and service time, which
don't line up with the latency/transfer split. Service time is the time
a warm system takes to process one request-response pair, while delay is
the time spent waiting for the cache to load, stuff get read from disk
and, especially especially, the time sitting in the run-queue, twiddling
your thumbs, waiting for a CPU.
Those aren't easily measured from outside the system, although I've
approximated them with a stepping load test. They're easy to measure
in-app, though.
Incidentally, did anyone notice that Intel revived their old “optimised for the
Internet” marketing fluff for their new Kaby Lake CPUs? Last seen for the
Pentium III, and just as disingenuous.
- Jonathan Morton
"Intel inside" is a warning label (;-))
--dave
--
David Collier-Brown, | Always do right. This will gratify
System Programmer and Author | some people and astonish the rest
dav...@spamcop.net | -- Mark Twain
_______________________________________________
Bloat mailing list
Bloat@lists.bufferbloat.net
https://lists.bufferbloat.net/listinfo/bloat
