WP switched from P e r l b a l to L V S
WP has switched from using P e r l b a l to L V S for Apache load balancing. Cut out a couple of hundred ms from each request and eliminated load oscillations on a ten to twenty second period. Too soon to tell how well it'll work but it's using less CPU and doing a nice balancing job. Not properly weighted yet - we have some new dual Opterons but mostly 3GHz P4s and the Opterons aren't weighted to give them more load.
Service times show the result. The change happened in two steps around 1700UTC on Sunday and that's visible still on the daily graph. The weekly shows the lower response time through the Monday peak. Not strictly comparable because one more database server is in operation but last week's Monday was typical and yesterday had higher transfer and request levels - around 400 megabits and 4,600 requests per second.
We're not using any of the enhanced features P e r l b a l offers, so it's not necessary to keep it to retain them.
As always, subject to change based on how well it seems to be working.
Service times show the result. The change happened in two steps around 1700UTC on Sunday and that's visible still on the daily graph. The weekly shows the lower response time through the Monday peak. Not strictly comparable because one more database server is in operation but last week's Monday was typical and yesterday had higher transfer and request levels - around 400 megabits and 4,600 requests per second.
We're not using any of the enhanced features P e r l b a l offers, so it's not necessary to keep it to retain them.
As always, subject to change based on how well it seems to be working.