RE: Network interface testing utlity
>---- Original Message ----
>To: email@example.com, firstname.lastname@example.org
>Subject: RE: Network interface testing utlity
>Date: Mon, 5 Oct 2009 23:37:14 -0700
>>> Give me an idea of what parameters (e.g. throughput, error rate)
>>> you intend to measure. The measured parameter will give us an idea
>>> of the measurement intervals.
>>I think throughput rate for sure. Also errors encountered. To
>summarize I would like to test the overall reliability (not sure how
>to quantify it) of the media types. My analysis does not have to be
>very very precise but a rough idea would be good enough.
>>I do blog at http://blogs.koolwal.net/
OK here's the short answers. The reliability of a "box" (such as a
network card) is usually measured as Availability (the fraction of
time the box is up and running correctly). The two parameters used
to calculate availability are MTBF (Mean Time Between Failure) and
MTTR (Mean Time to Restore). Since you haven't got lots of years to
measure, I'd suggest you replace any Availability measurements with
testimonials on reliability from users, satisfied or not. This is a
heuristic but IMHO it's better than anything you will get from a
The usual measure of media performance is the error rate, defined
either as the % of bits/blocks errored dividided by the total
bits/blocks transmitted or as a rate (errored bits/blocks per
second). These days with everything transmitted as blocks with CRCs
or checksums, the block error measurements make more sense. To
measure only the media, the best place to make such measurements is
at the link level (e.g. the Network Card) if that card can be made to
tell you when it detects block errors. If not proceed to plan c.
Probably the best measure of performance is Throughput, measured
either as a % (blocks received successfully divided by total blocks)
as in "99.5% throughput" or as a rate as in 56kB/s throughput.
Usually the latter is the better measurement since it includes delays
and retransmissions. Throughput is measured "as close to the
ultimate application as possible"-probably TCP in todays networks.
As a result you need to measure the number of good and bad segments
received per unit time. You should do some googling to see what
range of numbers to expect. I would also suspect that most network
monitoring software will give you the values you need to compute
throughput or compute it for you.
As to measurement time there are two complementary rules:
1. Measure as long as possible (since throughput is an average); and
2. Measure until your numbers do not change significantly
>>Hotmail: Free, trusted and rich email service.
>>To UNSUBSCRIBE, email to debian-user-REQUEST@lists.debian.org
>>with a subject of "unsubscribe". Trouble?