grid systems output indexing
I'm deploying a grid-like network performance monitoring system on our
network (within a large DoD site) using iperf and other OSS tools.
Based on all the data we're aggregating (throughput, loss, latency,
jitter), I'd like to derive an index number for a "weather" motif. This
is for users not familiar with network performance attributes.
We've done some pilot work using equal weight for the above-listed
attribute results. We simply normalize the results on a scale of 100
(using specific, appropriate target values for each attribute) and
average the results together. Then we say something like 100-85 is
sunny, 84-75 is partly cloudy, etc.
For instance:
15 minute results of 68Mbps throughtput/.03% loss/1.52ms latency/.12ms
jitter
80Mbps is maximum possible for 100/full connected client so 68Mbps =
68/80 = score of 85
2% loss is maximum acceptable so .03% loss = 2-.03/2 = score of 98.5
20ms latency is maximum acceptable so 1.52ms latency = 20-1.52/20 =
score of 92.4
... you get the point.
The problem is that using lots of sample point data over a period of
time tends to wash out variance in our results and we're left with a
network that *always* tends to be partly cloudy. While this points to
stability, I'd like to be able to somehow capture frequency and
magnitude of deviation from normal in our index. This leads me to my
questions (whew!):
1. Has anyone developed any applied algorithms or statistical
mathematics that can be used with network performance test results to
give a "realistic" aggregate index number? Ideally, the math should
provide some (but not too much) variation in the index number. The
scoring method seems solid, it's just the average data used in the
scoring is too intolerant of medium to low-level deviation.
2. What type of weights would you ascribe to each network attribute
(throughput, loss, latency, jitter)? I've been using equal weight, but
perhaps that's not quite right.
What do you think?
If this is an inappropriate question for the iperf users list, my
apologies. If so, anyone know of the right place to pose these questions?
Tate
--
C. Tate Baumrucker
Principal Architect
Callisma
301.938.3748 - cell
571-633-9500 - direct line
571-633-9788 - fax
tate.baumrucker --at-- callisma.com
http://www.callisma.com