| 2011-03-03 09:18:19 |
Robert Collins |
description |
We have an excellent serverside render time report, but we have no measurement or reporting on the total in-datacentre times.
This is needed to detect misconfiguration/overload/backups in the datacentre request path.
A low key approach would be to just measure the response times for urls in apache. Theres probably prepackaged things we could use to get some interesting reports on that.
A deeper (and more useful) analysis would be to split out the requests going to squid and those going to the appservers.
Deeper still would be to group urls to page ids using the zserver reference logs, and then generate the same time metrics - 99th percentile, requests, min, a graph etc.
A big initial step would be the 99th percentile apache completion time compared to the 99th percentile render time. |
We have an excellent serverside render time report, but we have no measurement or reporting on the total in-datacentre queuing / latency times.
This is needed to detect misconfiguration/overload/backups in the datacentre request path.
A low key approach would be to just measure the response times for urls in apache. Theres probably prepackaged things we could use to get some interesting reports on that.
A deeper (and more useful) analysis would be to split out the requests going to squid and those going to the appservers.
Deeper still would be to group urls to page ids using the zserver reference logs, and then generate the same time metrics - 99th percentile, requests, min, a graph etc.
A big initial step would be the 99th percentile apache completion time compared to the 99th percentile render time. |
|
