greenspun.com : LUSENET : ISP Quality-of-Service Monitor : One Thread

I notice you are polling a bit too early. I.e. the poll happened at 19:15:15 but the testing at that time did not finish until 19:16:34 so you end up getting the previous periods result.


-- uniken (uniken@optushome.com.au), May 15, 2000


Yeah, the latest version of the status script (0.4) actually doesn't use cron anymore, and just runs every 15 minutes from when you start it. The rationale for this was because if everyone pinged the same site at exactly the same time, the increased network load would mean inaccurate results.

So the latest version assumes that you start the script at some random time, resulting in a steady stream of small pings rather than a DoS-like massive ping attack on a server at one time.

Back to polling. The script that polls the connection reports run every 15 minutes via cron. If your test didnt finish until 19:16, the server will fetch those results at 19:30, and your 19:31 test results at 19:45 and so on.

I realise that this leads to a bit of a timing inaccuracy, but with 7 nodes, the average should result in a usable graph. The worst case scenario, of course, is that everyone tests at 19:16, and if the server gets the 19:15 results for everyone, the graph will be 15 minutes late.

So in essence we have to trade off time-base accuracy for being polite on the network, not hitting the servers in one big distributed ping flood type thing, and affecting the results/getting people pissed off.

If you have any other suggestions on how to do this post them in the forum, and I'll see what I can do to get them implemented.


-- Justin Chan (jc@iah.net), May 15, 2000.

Yes this is sensible and the way to go.


-- uniken (uniken@optushome.com.au), May 17, 2000.

Moderation questions? read the FAQ