edward@pic.ucla.edu (Edward Dergharapetian) (02/19/89)
Hi--
Does anyone out there know exactly how traffic(1) figures out what the
load on the ethernet is? I've looked at the ether(3) rpc calls and the
structures they return, but couldn't find any values indicating the load
(or percentage thereof). The code for traffic is almost impossible to
decipher. Is it making assumptions about the capacity of the ethernet? We
would like to use the same algorithm to monitor the load on our network,
but do it in batch (i.e.: through crontab or a daemon). Traffic is too
interactive for this purpose. Also let me know if such a tool already
exist. Please direct your responses to the e-mail address below. Thanks
for your help,
edward.
Edward Dergharapetian (213) 206-6067
UCLA Mathematics Dept. 405 Hilgard Ave. Los Angeles, CA 90024-1555
UUCP:.!{rutgers,ucbvax,sdcrdcf,{hao!cepu}}!ucla-cs!math.ucla.edu!edward
ARPA: edward@math.ucla.edu BITNET: edward%math.ucla.edu@INTERBITedward@math.ucla.edu (Edward Dergharapetian) (03/10/89)
Thanks to everyone who replied. Although some people had suggestions on
how traffic(1) works, alot of people asked me to post a summary of the
answers. Here is what I found after reading some of the replies and
further digging into the code. I originally thought of "load" monitoring
in traffic as being:
(packet count/size being broadcast)/(packets that the ethernet can handle)
As it turns out that is NOT what the "load" monitor in traffic does. What
actually happens is that traffic measures the number of packets that were
read off the ethernet during an interval. This is then divided by the
time interval in question. Some scaling is performed (i.e.: what you see
on the traffic screen is in bits/millisecond). So in essence traffic
measures the average throughput of the system during a certain time
interval. It cannot determine that the ethernet was flooded (i.e: %100
load), since it doesn't know how many packets the ethernet can handle (no
assumption is made). I guess the term "load" might not be appropriate in
this case.
Some people were also inquiring about any batch-like programs that collect
ethernet stats. A statistics collection program called nnstat is
available via anonymous ftp from 'isi.edu'. It seems pretty usable;
however, it too doesn't provide true load statistics.
edward.
Edward Dergharapetian (213) 206-6067
UCLA Mathematics Dept. 405 Hilgard Ave. Los Angeles, CA 90024-1555
UUCP:.!{rutgers,ucbvax,sdcrdcf,{hao!cepu}}!ucla-cs!math.ucla.edu!edward
ARPA: edward@math.ucla.edu BITNET: edward%math.ucla.edu@INTERBIT