Hyperthreading and CPU Hog Processes

[stultus]

Hypothetical question: I have a server with HT on and a process that occasionally hits 50% CPU usage (as seen through XP's task manager). The process is not SMP-capable or multithreaded, blah blah. Would I see a performance boost if I disabled HT? Or, is the process secretly taking what it needs from the virtual CPUs?

 

[DAPUNISHER]

There are some articles exploring HT over@2CPU Text Hit up the ones on the recent articles linked in the lower left. Should provide some illumination on the topic 🙂

 

[LTC8K6]

HT doesn't slow apps down so I wouldn't worry about it.

Unless you have one of the older Xeons back from when HT was first introduced. Those can have slowdown problems.

Does it show 50% use on both CPU's? ---- Hmmm......somehow I missed that big giant word Hypothetical. 😀

 

[iwantanewcomputer]

only way to know for sure with that app is to bench it yourself somehow. i think at worst hyperthreading will make something like 1% slower

 

[VirtualLarry]

Originally posted by: LTC8K6
HT doesn't slow apps down so I wouldn't worry about it.

Unless you have one of the older Xeons back from when HT was first introduced. Those can have slowdown problems.

Does it show 50% use on both CPU's? ---- Hmmm......somehow I missed that big giant word Hypothetical. 😀

I think that the issue in question is more related to how XP displays CPU usage, running on a HT-enabled machine. The display is not accurate. A task can be effectively taking up the full 100% of the physical CPU, yet will only ever display a maximum of 50% CPU usage, using an HT-enabled system.

Btw, have HT enabled can easily slow things down, relative to a non-HT rig, depending on what kinds of tasks need to be performed. AFAIK, the implementation of HT in the older Xeons, and in the Northwood HT-enabled desktop CPUs, are identical. As far as any improvements made in Prescott HT-enabled CPUs, I don't know. I know that there were rumors of improvements. But the bottom line is that the physical CPUs resources will always be shared while running under an HT-enabled configuration, and that there are degenerate cases of loads, that will cause thrashing of those shared resources, resulting in an overall loss of performance, relative to running those two tasks sequentually in a non-HT-enabled system. (Speaking of CPU resource requirements here, not system I/O resources, etc.)

 

[stultus]

I'm running a game server that happens to be a CPU hog. This is off a northwood P4. What's a good way to measure "real" processor usage? It will sometimes hit "50%", but if this isn't accurate I need a better way to measure.

 

[DAPUNISHER]

Originally posted by: stultus
I'm running a game server that happens to be a CPU hog. This is off a northwood P4. What's a good way to measure "real" processor usage? It will sometimes hit "50%", but if this isn't accurate I need a better way to measure.

Turn HT off and see how it does as you already reasoned out 🙂