Operating in a very separated IT-as-infrastructure/Engineering-as-Software environment.
I have a server that, after random periods of uptime, suffers from severe CPU speed degradation.
The server is a largely unmodified Apache webserver running on RedHat, using AMD Opteron(tm) Processor 6176 SE.
cat /proc/cpuinfo
over various periods will report roughly the spec'ed hardware speeds (2300MHz) when the server first restarts. Over time, this reported speed will degrade to ~600MHz. This is reflected in multiple processes slowing down (rendering, interpreter times, content delivery, etc). We cannot even stress these CPUs back to their limits in targeted fashion.
We've requested that our IT organization look closely at the hardware to make sure there is no failure going on. Last time they looked they determined a backup PSU was bad, but that doesn't explain why the server would degrade over time, does it? Since the PSU was replaced, we're still seeing the behavior on this server. Now that we're requesting further inquiries, IT has resorted to canned messages of "We've investigated the server and can find no issues," even while the CPUs are operating <30% of their advertised speeds.
Is there something obvious I'm missing? At this point it feels like I need to escalate this failure to management but I'm searching for all other options. I'm definitely not a SysAdmin, and don't understand what could be happening that's crippling my resources so terribly.
cpupower frequency-info
over time? (If the server's too old to have that, thencpufreq-info
instead.)