| View previous topic :: View next topic |
| Author |
Message |
martin.k
Guru
Joined: 28 Nov 2004
Posts: 493
Location: Wylatowo, Polska
|
 Posted: Thu Mar 16, 2006 6:39 pm Post subject: [KERNEL]Genetic anticipatory is back |
 |
|
Hi!
I think that some of you will be glad to hear that Jake Moilanen is back on genetic I/O scheduler hacking 
Now testing 2.6.15-ck7 + reiser4 + genetics
Works perfect!
More interesting stuff to come from Jake and IBM... Soon...
http://jakem.net
http://kernel.jakem.net
http://kernel.jakem.net/patches/genetic/2.6.15/
_________________
linux-2.6.17 +ck +R4 +lockless +genetic-as +... więcej nie pamiętam
LRU #299256 |
|
| Back to top |
|
 |
seren
Guru
Joined: 27 Aug 2005
Posts: 448
Location: Wisconsin
|
| Posted: Thu Mar 16, 2006 8:18 pm Post subject: |
|
|
| be nice to see who impliments this into their patchset. as i would really like to try it, maybe nitro-sources? maybe no-sources? |
|
| Back to top |
|
|
cheater1034
Veteran
Joined: 09 Sep 2004
Posts: 1558
|
| Posted: Thu Mar 16, 2006 8:29 pm Post subject: |
|
|
Hmm, I would be interested next stable release.
I could use all of them (zaphod, etc), but I wouldnt if it broke deadline  which it says it does now.
_________________
IRC!: #zen-sources on irc.rizon.net
zen-kernel.org
--
Lost in android development land. |
|
| Back to top |
|
|
martin.k
Guru
Joined: 28 Nov 2004
Posts: 493
Location: Wylatowo, Polska
|
| Posted: Thu Mar 16, 2006 8:46 pm Post subject: |
|
|
| cheater1034 wrote: | Hmm, I would be interested next stable release.
I could use all of them (zaphod, etc), but I wouldnt if it broke deadline which it says it does now. |
I have not tested deadline - but currently it does nothing to deadline. The 2.6.15-gl1 is only modding anticipatory scheduler.
P.S.
Jake's page hasn't been recently updated. Hope it changes soon.
_________________
linux-2.6.17 +ck +R4 +lockless +genetic-as +... więcej nie pamiętam
LRU #299256 |
|
| Back to top |
|
|
cheater1034
Veteran
Joined: 09 Sep 2004
Posts: 1558
|
| Posted: Thu Mar 16, 2006 8:58 pm Post subject: |
|
|
I don't think I'll put it in no-sources, unless it doesnt affect anything and is only a backup options.
Been researching the issue, havent used genetic before, the genetic algorithms is a great theory, but hearing the information about former debates, it adjusts too slowly and actually increases latencies a lot.
I don't know if it is the best choice
_________________
IRC!: #zen-sources on irc.rizon.net
zen-kernel.org
--
Lost in android development land. |
|
| Back to top |
|
|
martin.k
Guru
Joined: 28 Nov 2004
Posts: 493
Location: Wylatowo, Polska
|
| Posted: Thu Mar 16, 2006 9:15 pm Post subject: |
|
|
| cheater1034 wrote: | I don't think I'll put it in no-sources, unless it doesnt affect anything and is only a backup options.
Been researching the issue, havent used genetic before, the genetic algorithms is a great theory, but hearing the information about former debates, it adjusts too slowly and actually increases latencies a lot.
I don't know if it is the best choice |
O.K.
Just wanted to share that info with other guys here - not to persuade you on including it into your patchset.
_________________
linux-2.6.17 +ck +R4 +lockless +genetic-as +... więcej nie pamiętam
LRU #299256 |
|
| Back to top |
|
|
seren
Guru
Joined: 27 Aug 2005
Posts: 448
Location: Wisconsin
|
| Posted: Thu Mar 16, 2006 10:40 pm Post subject: |
|
|
i patched ck 2.6.15-ck7 with those patched and didnt find anything in make menuconfig to enable them, i just assume they will automagically work? im compiling the kernel now
thanks
james |
|
| Back to top |
|
|
martin.k
Guru
Joined: 28 Nov 2004
Posts: 493
Location: Wylatowo, Polska
|
| Posted: Thu Mar 16, 2006 11:14 pm Post subject: |
|
|
| seren wrote: | i patched ck 2.6.15-ck7 with those patched and didnt find anything in make menuconfig to enable them, i just assume they will automagically work? im compiling the kernel now
thanks
james |
You should have done make oldconfig first. Or:
| Code: | Library routines --->
[*] Genetic Library
-------------------------
Block layer --->
IO Schedulers --->
<*> Anticipatory I/O scheduler
[*] Genetic Anticipatory I/O scheduler (EXPERIMENTAL)
-------------------------
|
Genetic anticipatory Depends on: IOSCHED_AS && GENETIC_LIB && EXPERIMENTAL
So you need to set: Code maturity level options ---> [*] Prompt for development and/or incomplete code/drivers and then the above mentioned switches.
_________________
linux-2.6.17 +ck +R4 +lockless +genetic-as +... więcej nie pamiętam
LRU #299256 |
|
| Back to top |
|
|
cheater1034
Veteran
Joined: 09 Sep 2004
Posts: 1558
|
| Posted: Thu Mar 16, 2006 11:14 pm Post subject: |
|
|
| seren wrote: | i patched ck 2.6.15-ck7 with those patched and didnt find anything in make menuconfig to enable them, i just assume they will automagically work? im compiling the kernel now
thanks
james |
You can use it by the elevator= boot parameter, or make it default in Block Layer > Io Schedulers.
-----
Personally, I don't see the genetics increasing hard drive performance, Make sure you test it before using it, your latencies actually may be way higher than regular deadline or anticipatory.
_________________
IRC!: #zen-sources on irc.rizon.net
zen-kernel.org
--
Lost in android development land. |
|
| Back to top |
|
|
seren
Guru
Joined: 27 Aug 2005
Posts: 448
Location: Wisconsin
|
| Posted: Fri Mar 17, 2006 2:06 am Post subject: |
|
|
martin_k
how has the performance been for you?
cheater
have you tested it? |
|
| Back to top |
|
|
cheater1034
Veteran
Joined: 09 Sep 2004
Posts: 1558
|
| Posted: Fri Mar 17, 2006 4:44 am Post subject: |
|
|
| seren wrote: | martin_k
how has the performance been for you?
cheater
have you tested it? |
I haven't tested it recently. In the past there was a debate about this for love-sources. It was tested out with nicksched too, in call cases the genetic idea was the same, it actually increased latencies.
_________________
IRC!: #zen-sources on irc.rizon.net
zen-kernel.org
--
Lost in android development land. |
|
| Back to top |
|
|
Tiger683
Veteran
Joined: 08 Jan 2005
Posts: 1347
Location: Heffner's House
|
| Posted: Fri Mar 17, 2006 12:44 pm Post subject: |
|
|
Genetic algorithms in IO- or cpu-scheduler actually KILL your performance.
Any improvement is placebo, they're only suitable for solving linear/static problems,
which things like scheduling apparently aren't. Alone the calculations cause more overhead than gain.
So, -nitro won't touch it, feel free to do it in -no or whatever.......
_________________
Retired gentoo user |
|
| Back to top |
|
|
martin.k
Guru
Joined: 28 Nov 2004
Posts: 493
Location: Wylatowo, Polska
|
| Posted: Fri Mar 17, 2006 1:22 pm Post subject: |
|
|
Do as you like it...
I'm not going to force nor persuade anyone to use geneics.
I use it and just wanted to share some info with others.
_________________
linux-2.6.17 +ck +R4 +lockless +genetic-as +... więcej nie pamiętam
LRU #299256 |
|
| Back to top |
|
|
predatorfreak
l33t
Joined: 13 Jan 2005
Posts: 708
Location: USA, Michigan.
|
| Posted: Fri Mar 17, 2006 2:45 pm Post subject: |
|
|
| cheater1034 wrote: | I don't think I'll put it in no-sources, unless it doesnt affect anything and is only a backup options.
Been researching the issue, havent used genetic before, the genetic algorithms is a great theory, but hearing the information about former debates, it adjusts too slowly and actually increases latencies a lot.
I don't know if it is the best choice :) |
From what I know, it's better for FINDING very abnormal tuning settings, rather than desktop usage.
(e.g Find optimal tunables for a VERY insane workload and set them statically, so you don't have to recalculate this stuff over and over.)
_________________
System: predatorbox
Distro: Arch Linux x86_64
Current projects: blackhole, convmedia and anything else I cook up. |
|
| Back to top |
|
|
cheater1034
Veteran
Joined: 09 Sep 2004
Posts: 1558
|
| Posted: Fri Mar 17, 2006 4:49 pm Post subject: |
|
|
It does decrease performance on anything it is applied over, such as it was applied over nicksched and it killed it in performance.
I would apply it for the select few who would want to use it, if they need it for their system. I will never touch it with even a 10 foot pole if it changes the as or deadline code. If it creates a new file and leaves AS alone it will be acceptable.
_________________
IRC!: #zen-sources on irc.rizon.net
zen-kernel.org
--
Lost in android development land. |
|
| Back to top |
|
|
predatorfreak
l33t
Joined: 13 Jan 2005
Posts: 708
Location: USA, Michigan.
|
| Posted: Fri Mar 17, 2006 7:08 pm Post subject: |
|
|
| cheater1034 wrote: | It does decrease performance on anything it is applied over, such as it was applied over nicksched and it killed it in performance.
I would apply it for the select few who would want to use it, if they need it for their system. I will never touch it with even a 10 foot pole if it changes the as or deadline code. If it creates a new file and leaves AS alone it will be acceptable. |
I'm not suggesting you DO, I'm stating that it is nice for finding very abnormal tunables and then setting them STATICALLY and reversing genetic.
_________________
System: predatorbox
Distro: Arch Linux x86_64
Current projects: blackhole, convmedia and anything else I cook up. |
|
| Back to top |
|
|
martin.k
Guru
Joined: 28 Nov 2004
Posts: 493
Location: Wylatowo, Polska
|
| Posted: Tue Mar 21, 2006 8:34 am Post subject: |
|
|
There's an update of genetic-lib - 2.6.16-gl1
It introduces _experimental_ fingerprinting code which makes workload data to be classified and used to tune the I/O schedulers. The code is under _heavy_ development so beware
http://kernel.jakem.net/patches/genetic/2.6.16/
_________________
linux-2.6.17 +ck +R4 +lockless +genetic-as +... więcej nie pamiętam
LRU #299256 |
|
| Back to top |
|
|
seren
Guru
Joined: 27 Aug 2005
Posts: 448
Location: Wisconsin
|
| Posted: Wed Mar 22, 2006 6:27 pm Post subject: |
|
|
| i tried this patch and i got a kernel panic for the first time in three years ;/ i think it was because i run F@H, for all you F@H members dont try this patch, my assumptions are because when i removed foldingathome form init it started when i ran it craped on me bad.,, |
|
| Back to top |
|
|
martin.k
Guru
Joined: 28 Nov 2004
Posts: 493
Location: Wylatowo, Polska
|
| Posted: Thu Mar 23, 2006 12:18 pm Post subject: |
|
|
Too bad 
It compiles fine but ends with kernel panic after reboot. It's due to fingerprinting code. For now do not even try to enable it... 
At least my reiser hdd survived and still works 
_________________
linux-2.6.17 +ck +R4 +lockless +genetic-as +... więcej nie pamiętam
LRU #299256 |
|
| Back to top |
|
|
|
Unsupported Software
