1206211443 T * * http://linux-vserver.org/ |stable 2.2.0.7, devel 2.3.0.34, grsec 2.2.0.7|util-vserver-0.30.215|libvserver-1.0.2|vserver-utils-1.0.3| He who asks a question is a fool for a minute; he who doesn't ask is a fool for a lifetime -- share the gained knowledge on the Wiki, and we forget about the minute. 
1206211443 T * harry -
1206211616 J * doener_ ~doener@i577B9DD6.versanet.de
1206211668 N * DoberMann[PullA] DoberMann
1206212032 Q * doener Ping timeout: 480 seconds
1206212476 J * trippeh atomt@uff.ugh.no
1206212854 J * hparker ~hparker@linux.homershut.net
1206214163 Q * hparker Quit: Read error: 104 (Peer reset by connection)
1206214378 P * the-me Verlassend
1206215763 Q * kiorky Ping timeout: 480 seconds
1206215997 J * kiorky ~kiorky@cryptelium.net
1206216712 J * JonB ~NoSuchUse@0x57325fba.kjnxx10.adsl-dhcp.tele.dk
1206216827 Q * larsivi Ping timeout: 480 seconds
1206217235 Q * JonB Quit: This computer has gone to sleep
1206217427 J * JonB ~NoSuchUse@0x57325fba.kjnxx10.adsl-dhcp.tele.dk
1206217512 Q * JonB 
1206217859 Q * doener_ Quit: leaving
1206218023 J * liona29 ~liona29@d033.dhcp212-198-248.noos.fr
1206218071 J * JonB ~NoSuchUse@0x57325fba.kjnxx10.adsl-dhcp.tele.dk
1206218080 Q * JonB 
1206218139 Q * liona29 
1206218210 J * mick_work ~clamwin@h-74-2-196-226.miatflad.covad.net
1206221203 J * JonB ~NoSuchUse@0x57325fba.kjnxx10.adsl-dhcp.tele.dk
1206221637 J * Aiken ~james@ppp121-45-199-100.lns1.bne1.internode.on.net
1206221932 Q * opuk Quit: leaving
1206222373 M * etteyafed Are there any docs that tell how one might setup scheduling for vservers as described at http://linux-vserver.org/CPU_Scheduler ?
1206222401 M * etteyafed That would be exceptionally useful.
1206222587 M * JonB etteyafed: dud you read the wiki?
1206222595 M * JonB did you read the great flowerpage?
1206222596 Q * dna Quit: Verlassend
1206222616 M * etteyafed I read the wiki and am inm the process of looking at the GFP
1206222659 M * JonB ok
1206222676 M * etteyafed I thought the GFP was the 2.0 reference
1206222819 J * larsivi ~larsivi@144.84-48-50.nextgentel.com
1206222956 J * opuk ~kupo@c213-100-138-228.swipnet.se
1206223323 Q * mnemoc Ping timeout: 480 seconds
1206223801 J * mnemoc ~amery@kilo105.server4you.de
1206223940 M * etteyafed JonB Do you use the /etc/vservers/srever/sched/ dir?
1206223979 M * JonB i dont use scheduling at all
1206224082 M * etteyafed The documentation is there, but its not easy to understand. 
1206224100 M * etteyafed Well at least not if you plan on using it.
1206224243 Q * mrfree Remote host closed the connection
1206224286 Q * Blissex Remote host closed the connection
1206224298 M * JonB i'm sorry, i can not help you any more. Ask Bertl or daniel_hozac 
1206224323 M * etteyafed Yeah, I will next time they are around. They may be arround even now.
1206224330 Q * JonB Quit: This computer has gone to sleep
1206224690 N * Bertl_oO Bertl
1206224705 M * Bertl etteyafed: what's up?
1206224782 M * etteyafed Hey, how might I set up my /etc/vservers/server/sched/ dir to support a somewhat fair share 
1206224804 M * Bertl http://linux-vserver.org/CPU_Scheduler
1206224804 M * etteyafed between two vservers and the host on a 2 cpu system
1206224857 M * etteyafed Yeah I read that an it is pretty complicated, but I am thinking that R = 2 T = 6 might do it
1206224859 M * Bertl if you want complete fair scheduling of all resources, then the ebst is to assign idle time shares to both guest
1206224896 M * Bertl the host will always get priority/shares then, but you should be fine with that (I assume the host is not cpu intense)
1206224897 M * etteyafed Well I would like CFS but that looks like it might make my brain explode
1206224930 M * etteyafed The host is not intense, it wil ldo minimal file shareing and that's it.
1206224934 M * etteyafed and some DNS
1206224944 M * etteyafed but that again is minimal
1206224961 M * Bertl as I assumed
1206224995 M * etteyafed so to do fair share I just touch idle-time and set up the interval2 and rate2 files?
1206225016 M * etteyafed should they be the same as the rate one files or do I not need both?
1206225026 M * Bertl yep, and if you want to give a little basic stuff to your guests (in case the host is doing more than expected)
1206225032 N * DoberMann DoberMann[PullA]
1206225051 M * Bertl then you can set something like 1/4 or 1/6 for the R1/T1 values
1206225068 M * Bertl 1/4 will mean one quarten when the host is busy
1206225074 M * Bertl *quarter
1206225093 M * etteyafed ok, so that means that they get less tokens when there is no idle time to be had?
1206225102 M * Bertl exactly
1206225117 M * etteyafed or at least a different R/T
1206225121 M * etteyafed great
1206225133 M * Bertl that's the idea behind the different scheduler settings
1206225151 M * Bertl i.e. they switch from R1/T1 to R2/T2 when the host is idle
1206225230 M * etteyafed This is really an excellent product, I don't know why it isn't more supported by the distros.
1206225243 M * etteyafed You would think RH would be all over it.
1206225261 M * etteyafed I guess they like XEN and wasted CPU better
1206225274 M * daniel_hozac it wastes RAM more.
1206225277 M * daniel_hozac faaar more.
1206225304 M * etteyafed if you don't need hw virt then using it is a waste... period
1206225354 M * etteyafed Vservers don't waste ram though as far as I can see
1206225374 M * daniel_hozac my point exactly.
1206225434 M * etteyafed How many tokens should I need to start? Just a few right? 2.6GHz maybe 1000? they are allocated every few ticks anyway
1206225479 M * Bertl the start value is not really important
1206225488 M * Bertl more important is the min/max
1206225527 M * Bertl as it controls how many tokens a guest can accumulate and how large the hysteresis (deadtime) is, once it runs out of tokens 
1206225617 M * etteyafed min should be low for me I don't want a high latency factor, and ill set max lower also, I don't want surges either
1206225707 M * etteyafed fill-rate2 will be used when there is idle time, but will fill-rate also be used. I mean the rates aren't cumulative right? that's the idea I get from all this anyway
1206225767 M * daniel_hozac if the system is idle, R2/I2 are used. if not, R1/I1 are used.
1206225782 M * Bertl don't go below 10 or so, and IMHO it is advisable to keep the fractions above 5 or 10 too (numerator side)
1206225789 M * etteyafed I should just set a 1/1 ration for r2/t2 then really
1206225813 M * daniel_hozac the guest will be able to use 100% CPU even though you have 1/10.
1206225832 M * etteyafed yeah
1206225843 M * etteyafed but that's only if there is idle time
1206225848 M * daniel_hozac if you have two guests, one with 1/10 and one with 2/10, you'll get 33% and 67%.
1206225851 M * Bertl but you will waste precious virtual time :)
1206225908 M * etteyafed so you say I shouldn't set R less than 5-10
1206225909 M * etteyafed ?
1206225928 M * etteyafed I will have to set interval higher then
1206225934 M * Bertl nah, usually above 5 (for R1 at least)
1206225957 M * etteyafed that makes sense
1206225959 M * Bertl and yes, you set the interval a little higher
1206225982 M * Bertl what you usually want to avoid iare setups like 1/10 or so
1206225994 M * Bertl because then the guest will run and wait for 9 ticks
1206226014 M * Bertl which usually causes higher task switching (and the resulting overhead)
1206226026 M * Bertl for the idle time fractions it doesn't matter
1206226053 M * etteyafed so 10/30 should be ok
1206226072 M * etteyafed I read It is advantageous to smooth operation of the algorithm to make the interval as small as possible (or much smaller than the bucket size). You can in most cases simplify the fraction, such as changing 30/100 to 3/10.
1206226155 M * Bertl yes, that improves things for high responsiveness and/or demonstration purposes, but usually you want more cpu time to work (i.e. larger chunks of tokens)
1206226170 M * Bertl but YMMV
1206226298 M * etteyafed well I am trying out 10/30, I want a relatively responsive setup, and all out speed isn't that important, but I may get better results with something like 100/300 or 50/150. I will just have to do some testing
1206226405 M * etteyafed I am sure that 50/150 will be pretty quick
1206226420 M * Bertl depends on your HZ settings
1206226441 M * Bertl if you have HZ=100, that means that you'll get 50 tokens every 1.5 seconds
1206226473 M * Bertl if you have HZ=1000, it is every 150ms :)
1206226501 M * etteyafed Hz is just the actual speed of your cpu though right
1206226521 M * etteyafed I am something like 2,600,000,000 like most of us
1206226540 M * daniel_hozac no, HZ is the timer interrupt.
1206226544 M * etteyafed oh
1206226560 M * etteyafed how do I find/set that?
1206226573 M * daniel_hozac grep CONFIG_HZ .config
1206226652 M * etteyafed I have no such file
1206226671 M * daniel_hozac in your kernel tree?
1206226677 M * etteyafed ok that
1206226679 M * daniel_hozac how did you manage to build that?
1206226701 M * etteyafed I don't have the kernel sources locally for security reasons
1206226709 M * etteyafed not that it matters much anymore
1206226877 M * Bertl daniel_hozac: could be tickless kernel
1206227064 M * etteyafed My kernel gets the Frontline flea advantage
1206227187 M * etteyafed well you know I didn't build my kernel I use daniel_hozac's Centos/RHEL 5 kernel-devel-2.6.22.19-vs
1206227199 M * etteyafed well I mean kernel-2.6.22.19-vs
1206227219 M * Bertl ah, then everything should be fine and work out of the box :)
1206227237 M * daniel_hozac (except lm_sensors on AMD64)
1206227237 M * daniel_hozac :)
1206227273 M * etteyafed I don't need that in a server, but I could make I work if it was needed. I have external hw that monitors temps
1206227402 M * etteyafed I have found that vservers do play quite well with selinux and the combination opens up a whole new avenue of security options
1206227420 M * etteyafed and headaches if you don't watch it but its mostly good
1206227855 M * etteyafed before I go, or one of you do what is the deal with cpu-id and priority-bias                 
1206227869 M * etteyafed and bucket-id
1206227882 M * daniel_hozac cpu-id lets you limit the definition to a specific CPU.
1206227890 M * daniel_hozac bucket-id isn't used at the moment.
1206227906 M * etteyafed Ok, so cpu-id is just the cpu number like 0,1
1206227908 M * etteyafed ,2
1206227910 M * daniel_hozac priority-bias is like a guest-wide nice-value.
1206227910 M * etteyafed etc?
1206227918 M * daniel_hozac yes
1206227947 M * etteyafed the higher the number in priority-bias the less priority your guest gets?
1206227976 M * daniel_hozac no. it's an actual priority value.
1206228050 M * etteyafed ok, so if guest A.priority == 1 and B.priority == 2 then who gets priority?
1206228056 J * Infinito ~argos@201-2-78-56.gnace701.dsl.brasiltelecom.net.br
1206228070 Q * Infinito 
1206228081 M * daniel_hozac assuming that the tasks have equal priorities, B.
1206228088 Q * bonbons Quit: Leaving
1206228225 M * etteyafed Thanks guys, and continue being awesome. 
1206228528 T * * http://linux-vserver.org/ |stable 2.2.0.7, devel 2.3.0.34, grsec 2.2.0.7|util-vserver-0.30.215|libvserver-1.0.2|vserver-utils-1.0.3| He who asks a question is a fool for a minute; he who doesn't ask is a fool for a lifetime -- share the gained knowledge on the Wiki, and we forget about the minute. 
1206228528 T * harry -
1206228563 J * anna30 ~anna30@d033.dhcp212-198-248.noos.fr
1206228685 Q * anna30 
1206229365 M * Bertl etteyafed: you're welcome!
1206229398 Q * ftx_ Ping timeout: 480 seconds
1206229454 J * hparker ~hparker@linux.homershut.net
1206229909 Q * docelic Quit: http://www.spinlocksolutions.com/