At 11:01 PM 10/3/99 -0300, Eric Dahnke wrote:
>> On Sun, 03 Oct 1999, Eric Dahnke wrote:
>> > Someone will scold me for this post, but would appreciate any thoughts:
>> >
>> > A T1 would be ~ 80% utilized passing 22,000msgs/hr if the average msg
>> > size was 23K.
>> >
>> > Thx
>>
>> You're not taking into account how the router will handle the traffic.
You can
>> compress data at the router, which can give significant performance
gains. e.g.
>> using stac compression on a Cisco
>
>Thanks all for the responses. There was a wide variation in replies (due
>to subjunctivity of the post), but there were a few replies which seemed
>to say ya -- 23K, 22,000 msgs/hr = 80% T1 utilization -- that sounds
>about right. Can I assume I was more or less on with my original
>assumption.
The discussion on peak vs average is worth understanding as is the fact that
a link that is 80% utilized will suffer a lot of latency and has no room for
significant down time.
80% average utilization is way high in my opinion. In fact, anything over
40-50% is risky.
Consider your T1 in the same light as a busy freeway. Now consider that
freeway to be at 80% utilization.
If a freeway has 80% of the number of cars it can hold, is it running
smoothly or is it pretty congested? Does any sort of hiccup on a busy
freeway total destroy the traffic flow? Does it take forever to recover once
there is a crash of some sort? Does any minor change in the flow create a
huge change in the time it takes to transit the freeway?
Note that at a certain point, any sort of comms link (nee queue) actually
degenerates with load. Consider what happens when a link gets so congested
that socket connections time out and packets get retransmitted due to
timeouts. That actually generates more traffic per message than a relatively
idle link.
Ignoring the latency induced traffic, any sort of down time is going to have
a serious effect on a highly utilized link. Eg, at 80% utilization, if your
link goes offline for, say, 5 hours, it'll take another 38 hours before your
link goes below 100% (assuming it can do 100% in the first place).
What that means is that if your link goes down at 1am and comes up at 6am
and someone injects an email later that day at 23:99pm, that email will be
delivered at 6pm the following day.
If your only concern is getting a certain volume out then that's probably
ok, if you have some delivery latency goals then it's almost certainly a
problem.
Mark.
