My Linux Performance HOWTO

[o0zi]

It's not in its finished state, but tell me what you think of my Linux Performance HOWTO.

[bsh152s]

Pretty good read. I'd known about a few you listed, but others I never really thought of.

[Sepero]

Cool, thanks. I love stuff like this.

[Hypz]

I cant unzip it to even read it. It downloads as
linux performance howto.zip and when I go to unzip it I get
unzip: cannot find linux, linux.zip or linux.ZIP.
Any one else have the same problem?

[o0zi]

Here it is as a zip without spaces in the filename, that should fix your problem.

[bwkaz]

Originally posted by Hypz
I cant unzip it to even read it. It downloads as
linux performance howto.zip and when I go to unzip it I get
unzip: cannot find linux, linux.zip or linux.ZIP.
Any one else have the same problem? Or, put quotes around the filename (or use tab completion, which will put backslashes before the spaces for you).

[Sepero]

Hey, I downloaded your file in hopes of reading it later. Now, I come to find out that you saved it in Microsoft's "rich-text-format". Could you please post it in html format.

[superted]

Is one of the points "making linux recognize rich text format" ? :p

[o0zi]

Well, for all of you without OpenOffice or Abiword...

[hard candy]

Well, for all of you without OpenOffice or Abiword...

I can read with it WinXP and MS Ofice. :D

Thanks for the tips.

[Sepero]

thanks, that's better.

[azambuja]

Hey, shouldn't there be a way to contact the author or at least the author's name?

Just a thought.
Fernando Azambuja

[o0zi]

Perhaps I'm too modest :p

Anyway, my email is o0zi (at) hotmail.com

[Darkbolt]

I did the noatime and removed the tty's, aswell as the locale stuff..

Im noticing a fair amount more ram available :)

(most of the stuff I'd already done)

Something you may want to include is to recompile X yourself, and to use CFLAGS when compiling software :)

[o0zi]

OK, added some stuff to my HOWTO from various JustLinux threads.

EDIT: Credit is due to bwkaz for the section on combreloc.

[schwartz]

thank`s a lot!!

[jot-87]

This is great, I'm a linux newbie and information like this is fantastic. I'm just wondering; will any of this affect the computers security? I'm setting up a server for a small website and I want it to be as securte as possible (without taking away too many options).

[o0zi]

They shouldn't affect the security of your machine, AFAIK.
If you pick fast software with an exploit, then it'll compromise your machine, but that's very unlikely.

[dboyer]

interesting... i tried the hdparm tool... you seemed to recommend changing multicount and i/o as "saftely" steps...

redhat9 set my drive to 16 multicount already, and i was getting ~490MB/s throughput constantly (ran a couple times)...

put I/O support from 0 to 32 bit... and the throughput went up to 512... and then to 186... then up to 512... very sporadic...

weird...

actually, i went back to reproduce it, and im no longer seeing that affect... oh well...

thanks for the doc... very helpful... the openoffice thing? hot damn! yay! i had heard about that about 5 minutes prior on these boards... but its sure nice for someone to write it down...

finally... don't have to wait 45 seconds for a wordprocessor to open...

[leonpmu]

to add a line in LILO, I dont know squat about Grub

In the append line - idebus=66 this will push the ide bus speed to 66Mhz and not 33, this also helps throughput alot especially on slower system, as long as they have a 66MHz bus, which most PII and up have.

Just a tip

L

[o0zi]

Ah yes, I forgot that one. That'll be in version 1.1:D

[bwkaz]

Originally posted by leonpmu
In the append line - idebus=66 Please, DO NOT, DO NOT, DO NOT DO THIS!

This setting does not change your IDE bus speed! It changes the kernel's idea of your PCI bus speed, which should always be 33MHz unless you've overclocked everything.

If you've read the warning message that seems to lead people to change this setting, you will have seen this:

ide: Assuming 33MHz system bus speed for PIO modes; override with idebus=xx From the first bold word there, you should see that it's not actually your IDE bus; it's you system bus (PCI bus).

The second bit that's in bold ("for PIO modes") is the reason that many people don't ever have a problem making that change. The kernel only uses this value if the IDE drive drops from DMA mode back to PIO, because in PIO mode, the kernel has to schedule read and write requests manually (basically). To do that, it needs to know how fast the PCI bus can operate, because it obviously can't hand requests to the IDE controller faster than the PCI bus will let it.

If you do set this value to 66, and your drive drops back out of DMA mode, the kernel will start sending requests up to twice as fast as the IDE interface can actually receive them -- and you stand a very good chance of losing data.

OTOH, setting idebus to 66 won't even speed things up in the vast majority of cases, because in the vast majority of cases, the IDE interface isn't running in PIO mode. And the idebus setting is completely ignored.

If you want to use higher UDMA modes (e.g. UDMA4 = ATA66, UDMA5 = ATA100, or UDMA6 = ATA133), then use hdparm to turn them on. Or just configure your kernel when you compile it, to turn on DMA automatically; it'll then choose the highest mode that both your IDE controller and your drive(s) can support (it also has a blacklist, IIRC, of chipsets that claim to handle certain modes but don't work right).

I don't know where people keep getting this "use the idebus setting" idea from, but it is most definitely not a good thing to do...

[o0zi]

OK bwkaz, thanks for the advice. Everyone pay attention to what the moderator says:p

[Nuada Storm]

Originally posted by Sepero
Hey, I downloaded your file in hopes of reading it later. Now, I come to find out that you saved it in Microsoft's "rich-text-format". Could you please post it in html format.

Rich-Text-Format is not something of Microsoft's it is an open standard.

[o0zi]

Well my second attachment is in HTML, so everyone should be able to read that. Yes, I think RTF is an open standard.