IBM just released a tutorial detailing the process involved in transforming “out of the box” Red Hat OS into a stable, secure, and better performing system customized to individual needs. Its based on Red Hat 7.3 but most of the techniques and procedures discussed are equally applicable to other Linux systems. This is also related to a “Speed Start Your Linux Program” and CD.
Tuning Red Hat for Maximum Performance and Security
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
The performance difference between a default install and a well configured operating system is large enough that I am surprised you don’t see more documentation on the subject floating around. It seems all too often that tweaking is an after thought. Anyway, for those that don’t feel like creating an account, IBM provides the guide in a zip file. Uploaded here http://users.adelphia.net/~geek/l-redhat.zip
I really don’t understand why distributors don’t put more effort in using these sorts of instructions for their default install.. It could be so easy. (Sure, sometimes hardware works against this, but that could be solved by making tweaks hardware dependent in the distribution.)
Jim, thanks for posting a direct link. I’m already registered with IBM, but many won’t want the hassle.
The tutorial is a good starting place, but parts of it seem less researched.
For instance, hdparm: -c1 is OK, but for most users with modern EIDE, -c3 is better (32 bit plus enhanced sync). Nowhere is MaxMultSect (moving multiple disk sectors for each I/O interrupt) mentioned, but typically by default this is off. Most recent HDs can handle 16 sector moves, so -m16 is worth trying. And -u1, allowing the O/S to process other I/O while waiting on the disk to finish transfers. It’s hard to imagine how one could have tuned Linux boxen with EIDE and not mention these parameters.
The kernel advice is good. Another area worthy of focus is to optimize for your particular CPU – see http://homepages.paradise.net.nz/frankkru/susei686/optimize.pdf for more info, including details on recompiling Qt/KDE.
I agree on EXT3 – I’ve been using it for at least a year without a hitch. It’s on this laptop, and on production servers I manage.
Back to the IBM registration issue – IBM DeveloperWorks has some really great content and worth giving them your Hotmail address (a user name and email addy is ALL they ask for – try *that* on MSFT MSDN or Borland’s sites!). Recently they’ve had good PHP code optimization tutorials, usability stuff, a great intro to chroot’ing Apache, etc. I’ve been registered with them for a few years now, and they have never sent me *any* marketing emails.
> but for most users with modern EIDE, -c3 is better
I think this is the other way around: for modern EIDE chipsets -c1 is better!
There is no big performance difference between -c1 and -c3. According to hadparm’s manual page the only differnce is that with -c3 additional sync sequences are send to the disc. Actually -c3 has more overhead then -c1, which reduces performance slightly. Those sync sequences are needed by some (older) chipsets. Thus I think the recommended -c1 setting is okay. Don’t use -c3 if your chipset doesn’t need it.
Regarding the developer works registration: I am registered since somewhen last year and never ever received any spam from them. There are some interessting articles there which make a registration worthwhile.
Cheers,
Andreas Simon
Thanks for the info & link
Actually the comments about hdparm are somewhat out of date. In modern versions of Linux, if your hard drive controller is supported (specifically listed in the kernel IDE controller config section, not just generically supported), the kernel will autotune the disk. Changing the settings with hdparm will then DECREASE your performance. For example, my 20GB Maxtor disk on my SiS IDE controller gets better performance with a value of 4 (instead of sixteen) for the multicount setting. The kernel correctly configures this setting, while doing it manually would probably decrease performance.
Andreas may be right about -c3. At least on my IBM Thinkpad, c1 vs. c3 doesn’t make any measurable difference. But that’s just one (somewhat unconventional) disk controller.
And I suspect my Thinkpad is also related to Rayiner’s comments. RH 7.3 optimizes disk quite well during install, but I still got a substantial boost via hdparm tuning. Laptops don’t see as widespread use, and typically have unique subsystems.
The way to tell if any of this helps is using the -Tt switch to benchmark things.