I think that I’m going to start looking for a new hosting provider.
A robot over at media temple said:
This is a courtesy notice to inform you that your (gs) Grid-Server’s GPU (Grid Performance Unit) usage for the current cycle is projected to exceed the amount included with your purchased plan.
— Current period: (01/24/2007 - 02/23/2007)
— Projected overage is: (77.00GPU)
— Purchased amount is: (1000.00GPU)
I went to check their online stats, and now they are projecting an overage of 139 nebulously defined “Grid Processing Units.”
Why not file a “support” request?
Are there any hourly or per-request metrics for GPUs? Unless I have fine grained reporting, I might as well be trying random optimizations when attempting performance. I turned off the heaviest queries in my gallery2, but apparently GPU usage is still going up.
They did respond quickly:
Currently there is no monitored service that will track you GPU usage hourly or in real time. We are working for a similar solution in the near future. You do however get an average of about 23,000 hits per weekday.
GPU’s are cumulative over the month — if you sustain that many hits CPU cycles will eventually add up and your GPU usage will continue to grow.
Random optimizations are a good way to go. Keep in mind stuff like queries to your database, cpu cycles consumed eventually adds up to 1 GPU. A large majority of our customers don’t even reach the 1000 GPU mark.
Really, who cares about real time response? Just put process id into the logs, and provide another log of cpu usage by process id. That wouldn’t be trivially easy for a user like me to arrange, but if they want to bill me for usage, they should be able to account for that usage.
How about a formula?
c1x cpu seconds of processing time + c2y queries + c3z cpu seconds of query processing time
where c1, c2, and c3 are published constants.