PCLOS2010beta..benchmarks..!!

Pclos 2010 beta ( ??beta1) rips ..
http://www.phoronix.com/scan.php?page=article&item=linux_distro_fourway&num=1

How bout dat !
PCLOS, the little engine that could. The end users friend.

 

Thanks - nice test suite.
Having read some of the comments on the tests I see once again that I know nothing. Eg. Take this comment: "One thing missing from the article is which of the tested distros were running with KMS enabled and which ones weren't."
To me that seems to imply that those with the knowledge can willy nilly change from KMS to UMS.
How is this done ? Is it 'nomodeset' in the kernel command line option ?
Not that I am itching to do this for the little extra speed that enabling UMS implies. Just curious as I have not encountered this KMS vs. UMS stuff before.

 

Although I like Phoronix tests, sometimes they make it wrong. Not here specifically, but in general, they don't always do proper Design of Experiment.

You need to check low-high mixes before you can really judge the impact. For example, both onboard GPU and a proper one, 512MB ram vs. 8GB RAM, 5400RPM disk versus 7200RPM and so forth. And then, you can look for nonlinear components and really know which one is due to hardware and which one due to kernel and bundled distro-specific stuff.

Maybe Mandriva could do better in OA FPS, but it was capped by hardware? You need to take the hardware out of the equation, meaning you need to be faster than software.

Not easy to test and then do all those ANOVA stuff, but it gives you pretty good and accurate results. For example, how much 32-bit processor plays part here? How much it's disk? Or GCC?

Maybe the packaging of GCC is all that matters? Then, you merely need a different version to have it work fine, with no regard to actual distro. This way, you can also expose performance bottlenecks and help build faster distros.

Mrk

 

Mrk,
Can you explain that bfs factor that PCLinuxOS is using in their builds? What is that? Also, the difference between KMS and UMS that Ocky mentioned?

Thanks.

BTW, PCLinuxOS has released a beta 2. Go to DistroWatch for further details. They also are offering a Gnome version. Just follow the Release Announcement in the section about the new beta 2. I downloaded the Gnome version (which is beta 1). Thought I might try it out later this evening.

Later...

 

Frankly, I've not yet explored it, therefore I'll keep comments for after I've studied the technologies at hand.
Mrk

 

BFS = Brain F*ck Scheduler

No, I am not kidding, that's really what it stands for. Essentially it is an alternative scheduler for the Linux kernel which supposedly performs better on desktop workstations (i.e. it doesn't scale as well as the default CFS but that's the whole point anyway). The author has tried to get it included in the official kernel but Linus Torvalds has denied it, saying it is superfluous and has no benefits over the default scheduler. If you don't know what a scheduler is, you can Wikipedia it.

Now as for the benchmarks, they don't matter to me. I don't care about 32 bit distros. Why would I want to use 32 bit when 64 bit has been around since like 2003? Whenever PCLOS decides to get with the 21st century maybe I will give it another look.

 

BFS is the legacy of legendary Con Kolivas.

http://ck.kolivas.org/patches/bfs/bfs-faq.txt

 

Thanks, chronomatic, for the info and reply. .

And, thanks, linuxforall, for the link. Very interesting read. I understand the scheduler concept and process a lot better now. From my limited understanding I don't see why bfs is not included by default.

Again, thanks to both of you.

Later...

 

The scheduling algorithm is probably the most complicated part of kernel design. That's why you have the classic CFS, RR, FIFO, and new ones are hard to come by. You want some ultra-geeks with PhDs for that to prove in the long run, you're not ruining your efficiency and performance.

I need to read about BFS to see what it gives ...

Mrk

 

Reading some of the scheduler features:

"The major limitations of BFS will be that of scalability, as the separate
runqueue designs will have less lock contention as the number of CPUs rises."

"The quadrupling for NUMA
is a value I pulled out of my arse."

Well, not sure about that.
Home use might be good, business use ... don't know.

Mrk