Linux: DeskOpt, "Completely Unfair Scheduling"

That is how one differentiates Linux Server Kernel from a desktop one. Not by having different releases, but tools like this one to help tweak /proc parameters.

I thought that's what "nice" was for.

Quoting:I thought that's what "nice" was for.

I was not [edited] aware that "Nice" collected statistics and gave recommendations, does it?

Any how, if it does, more sophisticated tools are needed to do a better job on collecting info over a long period of time, at least for server tuning.

How does "nice" adjust filesystem buffer sizes and network parameters? These two categories are the first place I would look to tune differently for desktop and server setups.

The "desktop" vs. "server" has been touted heavily during the scheduler argu^H^H^H^Hdiscussions recently.

Quoting:The "desktop" vs. "server" has been touted heavily during the scheduler

That is why I posted Bob. The script mentioned in the article is what I was debating for in the thread you mention. Other posters were thought having two different releases of the kernel instead. In my opinion, that is not necessary.

Such tools could tune the performance of the kernel by tailoring parameters based on statistics collected over a period of time. The tailoring will also take the hardware resource available in consideration. I didn't come up with this idea, it was a very helpful implementation on the VAX/VMS for a very long time.

>I didn't come up with this idea,

I used to be a perfomance guy in mainframe shops. We could tweak those old machines to within an inch of their lives, an important consideration for $100 million dollar systems.

I cannot do with Linux what I used to do with mainframes, however, because Linux does not offer anywhere near the control that the old mainframe systems did.

That, I think, is the problem:

Are adequate tools/settings available in Linuxland (or Unixland) to do the job adequately well? I don't think they are, but, as always, I could be wrong. I would be delighted to learn more than I know.

Quoting:I cannot do with Linux what I used to do with mainframes, however, because Linux does not offer anywhere near the control that the old mainframe systems did.

That, I think, is the problem:

Linux might not have what mainframes did, on the other hand, Linux hasn't been around as the long as the mainframe.

Also, recent Linux is more flexible and sophisticated than it has been, especially after IBM joined efforts with Red Hat & Suse to improve Linux for their main frame. I expect that to continue and become more valued technique to get the best out of expensive hardware.

Parameter tuning is more effective and valued on the server than desktop. Mainframe operations tend to be more consistent in their functionality, desktops are more erratic since they are more interactive event driven.

>since they are more interactive event driven.

Bingo.

OT:

Abe, I noticed you edited your first comment here without losing the formatting. How did you do that? I tried both with and without HTML, and I still lost all my paragraph breaks.

Edit: Testing Abe's following comment.

Quoting:I noticed you edited your first comment here without losing the formatting.

I did nothing special. Keep in mind that when you edit, you lose formatting during preview only. When you continue, all the formatting "magically" appear. I just ignore the preview state when editing.

Yup, OK. I was afraid to edit comments earlier, because the preview was busted.