1178411358 M * brcc_ Daniel, you still there ?
1178411381 M * brcc_ I am comparing vserver memory limits with a real linux server and things seems kinda weird
1178412078 M * fatgoose how weird?
1178412244 M * brcc_ i am doing some calculations here.. comparing a 128mb physical with a 128mb vserver
1178412596 M * fatgoose ok
1178412627 Q * yarihm Quit: This computer has gone to sleep
1178412820 M * Bertl brcc_: and what is weird?
1178412946 Q * bzed Quit: Leaving
1178413277 M * brcc_ bertl
1178413286 M * brcc_ Comparing with uml
1178413293 M * brcc_ it seems i can run much more stuff using uml
1178413298 M * brcc_ with same ammount of ram
1178413332 M * Bertl that would really surprise me :)
1178413369 M * brcc_ I am getting crazy over here, doing lot of comparisions. With vs2.0 if you ran 50 apache processess, they wouldnt be accounted
1178413373 M * brcc_ with vs2.2, 128mb ram, with 40 you run into problems
1178413428 M * Bertl and with uml?
1178413445 M * Bertl btw, do you have the limit overview of vs2.0 at hand?
1178413465 M * brcc_ Yes i have everything here
1178413471 M * brcc_ The limit when running 40 proccess ?
1178413480 M * brcc_ i mean, 100
1178413490 M * Bertl no, the vs2.0 with 50 or so
1178413504 M * Bertl or 100, whatever
1178413567 M * brcc_ Ok hold
1178413863 M * brcc_ Bertl: http://paste.uni.cc/15239
1178414039 M * Bertl and with 2.2 ?
1178414122 Q * gerrit Ping timeout: 480 seconds
1178414473 M * brcc_ http://paste.uni.cc/15240
1178414604 M * Bertl and you are sure you are running the same guests on them?
1178414646 M * Bertl e.g. the 2.0.2 has only half the number of files open
1178414653 M * brcc_ no
1178414656 M * brcc_ this guests are different
1178414659 M * brcc_ the same were the ones with java
1178414711 J * gerrit ~gerrit@c-67-160-146-170.hsd1.or.comcast.net
1178414751 M * brcc_ but what i mean is, on vs2.0 if you run 10 or 100 you dont lose much ram
1178414758 M * brcc_ on vs2.2 if you run 40 you lose almost 128mb
1178414783 M * brcc_ Imagine how sad would it be to upgrade a server from vs2.0 to vs2.2 without knowing this
1178414784 M * brcc_ hehe
1178414899 M * Bertl how do you define/test 'lose ram'?
1178414986 M * brcc_ running 100 apache proccess would use around 50mb
1178415001 M * brcc_ running 120 would use around 60-70 (just examples)
1178415006 M * brcc_ (this on vs2.2)
1178415010 M * brcc_ oops
1178415014 M * brcc_ this on vs2.0
1178415020 M * brcc_ while on vs2.2 40 would fit 128mb ram
1178415121 M * Bertl and you define 'use' as the RSS accounting?
1178415141 M * Bertl or do you do some serious checks on the host?
1178415180 M * brcc_ rss accounting, as shown in free -m
1178415199 M * Bertl okay, as we told you before, that was wrong/incomplete with 2.0
1178415202 M * brcc_ and doesnt make it possible to launch new proccesses when limit is reached
1178415207 M * brcc_ yes i got it
1178415208 M * brcc_ hehe
1178415241 M * Bertl so, except for the 'wrong' display in 2.0 (which we know about) nothing changed, no?
1178415277 M * Bertl btw, you can get the 'new' rss amount by adding up anon and rss (on 2.0)
1178415333 M * Bertl what seems more important to me would be to measure the memory consumption on the host
1178415340 M * Bertl (i.e. with /proc/meminfo)
1178415347 M * brcc_ yes i got it
1178415350 M * brcc_ new rss = old rss + anon
1178415355 M * Bertl and to compare that with e.g. UML
1178415387 M * Bertl if it turns out that UML uses less memory than Linux-VServer, then we have something to look into :)
1178415403 M * brcc_ But what i am afraid about is, with vs2.0 new apache proccesses could be forked without problems when new connections were received (until reaching maxclients)
1178415416 M * brcc_ Now this number of connections are really limited
1178415433 M * brcc_ maybe i am seeing it wrong
1178415450 M * Bertl well, raise the limit, or completely disable it
1178415475 M * Bertl your old setup has no limits on anon memory
1178415501 M * Bertl if you 'just' set a limit on RMAP, then you get the same behaviour (regarding limits) you had before
1178415516 M * Bertl (which I would not suggest to do, btw :)
1178415524 M * brcc_ hehe ghot it
1178415539 M * brcc_ on the old approach i had prople using, for example, 500mb but just showing 100mb. right ?
1178415551 M * Bertl something like that
1178415589 M * Bertl the values are not that critical, because if you have several guests, they might actually share the RMAP to some degree
1178415596 M * Bertl (not the ANON though)
1178415600 M * brcc_ i am wondering if i can run 100 proccesses on uml
1178415640 M * brcc_ Got it. so anon=shared
1178415681 J * er ~sapan@pool-71-168-215-87.cmdnnj.fios.verizon.net
1178415691 M * Bertl no, anon is anonymous mappings :)
1178415713 M * er hi Bertl
1178415725 M * Bertl hey er! LTNS!
1178415751 M * brcc_ ahh ok
1178415753 M * brcc_ sorry i read it wrong
1178415756 M * brcc_ the rmap not the anon
1178415757 M * brcc_ hehe
1178415757 M * Bertl np
1178415760 M * brcc_ too late here
1178415764 M * brcc_ going to go out and drink a bit
1178415766 M * er yep, i disappeared into a black hole, one of the many that I frequent
1178415770 M * brcc_ and think about memoryu limiting :)
1178415776 M * brcc_ thanks for the explanation bertl
1178415779 M * Bertl er: and how was it?
1178415780 M * brcc_ good night everyone
1178415786 M * Bertl brcc_: have a good one!
1178415867 M * er Bertl: not too bad, I've been in worse. 
1178415895 M * er Bertl: did you follow my thread with Daniel a little while ago?
1178415921 M * Bertl not much ...
1178415940 M * Bertl but I can read it up if it was important?
1178415976 M * er i can re-ask in 1 line
1178416122 M * Bertl then please do so :)
1178416192 M * er ok...if I modify sk->sk_xid for a connection from X->Y, will it get transferred to guest Y?
1178416248 M * Bertl the socket? no, only partially, i.e. it will give 'funny' results
1178416273 M * Bertl sockets currently carry 4 identifier, two for the xid and two for the nid
1178416450 M * er hm. ok. and is there a way of doing this?
1178416489 M * Bertl what do you actually want to do? move sockets around?
1178416515 M * Bertl and what about the tasks using those sockets?
1178416532 J * fatgoose_ ~samuel@206-248-175-36.dsl.teksavvy.com
1178416582 Q * fatgoose_ 
1178416597 M * er actually want to do - get a transparent HTTP proxy which works with VNET working without VNET
1178416650 M * er it consists of a process sitting in root context, which listens for HTTP requests, and depending on the Host: header, sets the xid of an incoming connection to the guest which has registered for a particular host
1178416716 M * Bertl why would you do that?
1178416742 M * er currently, the only 2 steps that VNET seems to carry out to this effect is to allow uspace processes to set a socket's xid
1178416767 M * er why: because guests share the same IP address
1178416786 M * Bertl yeah, but that is self inflicted, isn't it?
1178416791 M * er and their http servers need to run transparently
1178416795 M * er Bertl: I know:)
1178416806 M * Bertl and you need to use separate ports to bind to anyways
1178416826 M * Bertl so your transparent proxy has to map to various ports
1178416844 M * Bertl which in turn makes the context tagging obsolete
1178416871 Q * fatgoose Ping timeout: 480 seconds
1178416906 M * Bertl or am I missing something?
1178416981 M * er hm. VNET's autobind feature, which reroutes sys_bind seems to allow multiple guests to bind to the same port
1178417041 M * Bertl which sounds like a bug, and is a feature we didn't know about last time, no?
1178417062 M * Bertl how do you handle two sshd daemons bound to the same port?
1178417080 M * er right, because we weren't looking at anything other than raw sockets.
1178417115 M * er Bertl: VNET introduces an xid field in struct skbuff
1178417145 M * Bertl we use the fwmark or the secmark instead
1178417172 M * er and sets that field based on its own bind table in one of the NF hooks it implements
1178417176 M * er right
1178417199 M * er but right now you're only looking at those fields in RAW sockets.
1178417220 M * Bertl so how does it handle the two sshds bound to port 22?
1178417454 M * er I suppose it relies on the fact that when one of the bound guests accept()s, only one get_tcp_connection in tcp_ipv4.c will succeed - and that'll be the one with the right xid set for the connection
1178417481 M * Bertl okay, but which xid will be set?
1178417571 M * er Bertl: the transparent proxy defines one policy for that
1178417597 M * Bertl transparent proxy for ssh?
1178417623 M * er another policy is defined by libproper, which defines the slice which has access to the port at a time
1178417646 M * er Bertl: well, ssh makes this particular function a bug more than a feature
1178417681 M * er HTTP slides it closer towards a feature, even though the bugginess is quite clearly visible
1178417748 M * er anyhow, that seems to answer my question, you think that doing what VNET does right now should produce funny results
1178417764 M * Bertl well, if you want to implement something hacky like this, the best way is probably to add checks for the skb_mark to normal sockets, and disable the collision detection (or limit it to contexts)
1178417787 M * er right - I was planning to come to that eventually
1178417791 M * er VNET does the latter right now
1178417798 M * er and uses xids for the former
1178417801 M * Bertl so that ports can be bound without ever knowing that there is another guest using the same port
1178417824 M * Bertl but you should base that on nid instead of xid
1178417829 M * er ok
1178418054 Q * infowolfe Quit: Leaving
1178418843 Q * ensc Killed (NickServ (GHOST command used by ensc_))
1178418853 J * ensc ~irc-ensc@p54b4ecf5.dip.t-dialin.net
1178419296 J * DoberMann_ ~james@AToulouse-156-1-176-30.w90-38.abo.wanadoo.fr
1178419406 Q * DoberMann[ZZZzzz] Ping timeout: 480 seconds
1178419946 Q * FireEgl Quit: Bye...
1178419975 J * FireEgl FireEgl@2001:5c0:84dc:0:81bf:9202:bf7b:c83f
1178420158 Q * FireEgl 
1178420660 M * Bertl  okay, off to bed now .. have a good one everyone!
1178420667 N * Bertl Bertl_zZ
1178420887 J * ktwilight_ ~ktwilight@112.82-66-87.adsl-dyn.isp.belgacom.be
1178421297 Q * ktwilight Ping timeout: 480 seconds
1178422329 P * er 
1178423492 Q * cehteh Ping timeout: 480 seconds
1178425581 Q * shedi Ping timeout: 480 seconds
1178426116 J * shedi ~siggi@ftth-237-144.hive.is
1178433040 J * tudenbart ~willi@xdsl-81-173-229-65.netcologne.de
1178433491 Q * dothebart Ping timeout: 480 seconds
1178436261 J * phedny ~mark@ip56538143.direct-adsl.nl
1178436293 J * dna ~naucki@211-231-dsl.kielnet.net
1178436664 Q * phedny_ Ping timeout: 480 seconds
1178436916 Q * hardwire Ping timeout: 480 seconds
1178437855 J * infowolfe ~infowolfe@67.164.195.129
1178438718 J * mattzerah ~matt@121.50.222.55
1178442216 J * FireEgl FireEgl@Sebastian.Atlantica.CJB.Net
1178442599 J * infowolfe_ ~infowolfe@67.164.195.129
1178442921 Q * infowolfe Ping timeout: 480 seconds
1178443801 J * kwowt69 ~quote@pomoc.ircnet.com
1178443802 M * kwowt69 hi
1178443805 M * kwowt69 derjohn here?
1178443838 M * kwowt69 or anyone else, using his IP patch
1178444001 M * kwowt69 i've patched the kernel but i still dont know how to add more then 16 ips
1178444138 M * FaUl iirc you have to patch the util-vservver as well
1178444171 M * kwowt69 with which patch?
1178444495 M * FaUl i'm not quite sure
1178444513 M * FaUl i've never used that ip-patch
1178444565 M * kwowt69 argh
1178444581 M * kwowt69 how come every time i start to screw around with these IPs my filesystem breaks down!
1178444617 Q * bavi 
1178445140 J * yarihm ~yarihm@84-74-20-183.dclient.hispeed.ch
1178445191 N * DoberMann_ DoberMann
1178445920 M * sid3windr because you're using reiserfs?
1178446281 Q * sladen Ping timeout: 480 seconds
1178446404 M * daniel_hozac kwowt69: just add the addresses. util-vserver 0.30.212+ doesn't have a limit on the number of IP addresses.
1178446515 J * sladen paul@starsky.19inch.net
1178447042 J * ema ~ema@rtfm.galliera.it
1178447052 Q * Aiken Quit: Leaving
1178447773 M * derjohn kwowt69, yes
1178447836 M * derjohn kwowt69, I see daniel_hozac did alreadx answer. Besides that: it's not my patch, I just adopted what I found ....
1178447893 M * sid3windr adopted ? then it IS your patch! ;)
1178448346 Q * Nam Ping timeout: 480 seconds
1178448798 Q * infowolfe_ Quit: Leaving
1178448799 Q * derjohn Remote host closed the connection
1178448834 J * infowolfe ~infowolfe@67.164.195.129
1178449044 J * derjohn ~derjohn@80.69.41.3
1178449231 Q * gerrit Ping timeout: 480 seconds
1178449259 M * derjohn well, yes, I mean I "took that patch over into my patchset". But the patch is simply changing a single define. thats all .. (with .212+ utils you dont even have to change the utils anymore ...) I wonder when this define will become a compiletime option ...
1178449422 J * suser ~tux@bzq-82-81-19-143.red.bezeqint.net
1178449487 M * daniel_hozac make the patch?
1178449529 J * bzed ~bzed@dslb-084-059-121-166.pools.arcor-ip.net
1178449896 M * kwowt69 sid3windr: i'm using ext3
1178449965 J * gerrit ~gerrit@67.160.146.170
1178450399 Q * suser Remote host closed the connection
1178450522 Q * toidinamai_ Ping timeout: 480 seconds
1178451360 Q * yarihm Read error: Connection reset by peer
1178451469 J * yarihm ~yarihm@84-74-20-183.dclient.hispeed.ch
1178452335 Q * yarihm Quit: Leaving
1178452358 M * derjohn Bertl_zZ, daniel_hozac, bonbons: I changed in include/linux/percpu.h the  #define PERCPU_ENOUGH_ROOM to a value of 65536. Now everything seems to work, no more "could not allocate percpu ....". What it the major drawback of that? +32K seems not much nowadays.
1178452587 J * Piet hiddenserv@tor.noreply.org
1178452642 J * yarihm ~yarihm@84-74-20-183.dclient.hispeed.ch
1178453329 N * Bertl_zZ Bertl
1178453333 M * Bertl morning folks!
1178453407 M * Bertl derjohn: that is basically harmless, but the question is, does it work with mainline (which uses 32k) if so, the calculation must be going wrong somewhere and/or something uses more percpu space than expected (which should better be identified)
1178453454 M * Bertl derjohn: fact is, vs2.2.0 should precisely calculate the percpu room for the given number of (max) contexts
1178453599 M * Bertl okay, off for now .. back later ...
1178453603 N * Bertl Bertl_oO
1178453655 M * derjohn Bertl, hmmm, I greped though the sources for alloc_percpu and can only see ipv{4,6} stuff, vserver stuff and xfs_mount using it. I cannot say which module "combinations" show the problem, but I wonder why the problem does not come up with a AMD64 optimized kernel.
1178453706 M * derjohn bonbons says, he uses his patch with a similar patchset on i386 without probs .. but I didnt get him here for further investigation-
1178453881 M * derjohn hm, I reduced the # of max. CPUs vom 8 (default) to 4. Could that be the source od the prob? Well, I would think, I leaves more space over per cpu, but maybe the calculation does work "better"  with 8+ cpu?
1178453976 J * doener ~doener@host.magicwars.de
1178455924 J * Val ~val@v41.org
1178455933 M * Val hi :)
1178456087 M * Val kernel 2.6.20 vs 2.2.0 -> building one vserver with 1G space available dlimits set -> df -h show 100% (1Go) but only use 235M on disk... 
1178456128 M * Val an idea ?
1178456145 M * Val (util-vserver 0.30.212)
1178456164 M * Val (xid tags are set)
1178456341 M * derjohn Val, hm, you say within the guest there is 100% used, but from the host ( du -hs) only 235M are used?
1178456375 M * Val yes
1178456398 M * Val i see dlimit cache dir link are not done with vserver build command
1178456402 M * Val right ?
1178456425 M * Val testing again
1178456437 M * derjohn hm, did you delete stuff from the host ?
1178456489 M * derjohn did you set the path correctly ? (i.e. if you copy a /etc/vservers/xxx dir, the path which is limited still points to the old one)
1178456513 M * Val i did not copy anything
1178456522 M * Val i did build a new vserver
1178456536 M * derjohn Val, I assume you already did "chxid" ob the vserver's dir ?
1178456548 M * Val using an initpost script to put some missing stuff
1178456558 M * Val derjohn: yup :)
1178456579 M * derjohn hm, the % used up stuff should be updated immediately.
1178456589 J * bonbons ~bonbons@158.64.111.14
1178456603 M * derjohn Val,  maybe you also limited inodes ? maybe they ran put ?`
1178456605 M * derjohn *out
1178456608 M * Val perhaps the missing dlimits cache link is the trick
1178456610 M * derjohn bonbons, Weclome !
1178456633 M * Val derjohn: nope, 40000 i-nodes are set
1178456648 M * derjohn Val, I am usiing dlimits  but never had to link the cache (debian's util-vserver)
1178456662 M * Val i'm on debian too
1178456663 M * derjohn Val, 40000, well, that is not very much.
1178456680 M * Val ... this is the first time my script fail building a new vserver
1178456690 M * derjohn I had to rise several times for s.o. you compiles packages on the guest  (hello s0undt3ch !)
1178456705 M * derjohn Val, you had limits before ?
1178456715 M * Val yep, many times
1178456751 M * Val waityes it works
1178456758 M * Val it was the link cache
1178456764 M * Val :)
1178456765 M * derjohn hm, did you upgrade utils ?
1178456768 M * Val yep
1178456788 M * derjohn if so, could you file a bug or inform micah about what you found out ?
1178456794 M * Val what i dit : 
1178456799 M * Val mkdir -p /var/cache/vservers/dlimits/$vhost/cache
1178456805 M * derjohn (maybe daniel_hozac , too )
1178456825 M * derjohn Val, and this wasnt created with the .212 (etch?) utils ?
1178456828 M * Val derjohn: i will
1178456830 M * bonbons hey derjohn
1178456836 M * Val yes etch one
1178456846 M * derjohn Val, good to know ;)
1178456865 M * Val i'll publish my script too
1178456866 M * derjohn bonbons, hello! did you get my mail? If have one "update"
1178456871 M * derjohn Val, fine !
1178456889 M * derjohn Val, on the linux-vserver.org wiki, right ?
1178456893 M * Val when i'll take some time to patch vserver debootstrap build method to include it
1178456916 M * derjohn bonbons, can you read the backlog here? if not, I p-msg you
1178456949 M * derjohn Val, what do you want to patch there ?
1178456974 M * derjohn debootstrap is not responsile for the missing link 
1178456989 M * derjohn *responsible
1178456992 M * bonbons I have no backlog of today, but should be archived on l-v.org...
1178456994 M * Val some bad init scripts for devices / mount points etc.
1178457010 M * derjohn bonbons, i already p-msged you ...  got it?
1178457012 M * Val some rights/default xid/default limits
1178457046 M * Val improved autofs support for roaming users account between vservers
1178457048 M * Val ...
1178457076 M * derjohn hm, maybe it makes more sense to use the templaate funktion ?
1178457083 M * derjohn of util-vserver
1178457251 M * Val yup
1178457280 M * Val may be an addon to util-vserver package
1178457287 M * Val util-vserver-addon
1178457292 M * Val :)
1178457699 Q * gerrit Ping timeout: 480 seconds
1178457869 M * daniel_hozac Val: hmm? what was the problem?
1178457925 M * daniel_hozac the disk limit stuff works fine for me...
1178457967 M * derjohn daniel_hozac, the prob was a missing cache dir (see above)
1178458426 J * gerrit ~gerrit@c-67-160-146-170.hsd1.or.comcast.net
1178459124 N * DoberMann DoberMann[PullA]
1178460578 N * DoberMann[PullA] DoberMann
1178460640 P * Val 
1178460669 M * daniel_hozac derjohn: but that directory should be created by the tools...
1178460707 M * derjohn daniel_hozac, ok, the Val did something strange ... or maybe set a dlimit to that dir ;)
1178460758 M * daniel_hozac the host cannot be limited by disk limits.
1178461435 J * cehteh ~ct@pipapo.org
1178463130 J * transacid ~transacid@transacid.de
1178463654 Q * yarihm Ping timeout: 480 seconds
1178466171 J * toidinamai ~frank@i59F7192E.versanet.de
1178466627 J * yarihm ~yarihm@84-74-20-183.dclient.hispeed.ch
1178466685 Q * ema Quit: leaving
1178466872 M * s0undt3ch yep, I break all yor guests derjohn :)
1178466955 J * jrsharp ~jrsharp@c-71-228-234-243.hsd1.tn.comcast.net
1178466989 M * jrsharp hey all
1178466999 M * nebuchadnezzar hello
1178467008 N * Bertl_oO Bertl
1178467015 M * Bertl greetings!
1178467018 M * jrsharp anyone running vserver on the latest stable, 2.6.21.1?
1178467039 M * jrsharp I tried the latest posted patch, but it would not compile...
1178467039 M * nebuchadnezzar I just have a craisy GCC in a guest, I can not kill it, it just eat all my CPUs :-(
1178467055 M * jrsharp should I just be satisfied with 2.6.20.4?
1178467072 M * Bertl hmm?
1178467091 M * Bertl jrsharp: could you upload the compile error you get?
1178467096 M * jrsharp sure
1178467116 M * Bertl and what patch precisely did you use?
1178467175 M * brcc_ Bertl good morning
1178467182 M * brcc_ I am already wakeup and playing with memory limits
1178467208 M * brcc_ Using openvz i can have 200 apache proccess with memory usage which seems like vs2.0 behavior
1178467246 M * Bertl well, they probably get the accounting wrong too as we did before
1178467267 M * brcc_ hhehhehe
1178467270 M * Bertl but as I said, you should double check with memory usage on the host
1178467283 M * jrsharp bertl: the latest stable vserver patch for 2.6
1178467288 M * jrsharp it was meant for 2.6.20.4
1178467300 M * Bertl jrsharp: why not use the patch for 2.6.21?
1178467307 M * jrsharp I didn't see it
1178467319 M * brcc_ If i limit just rmap (and not rss), the free comand inside the guest would act weird, wouldn't it ? I mean, there would be memory problems when the free command still shows much memory ?
1178467347 M * jrsharp the latest listed on the website, and the latest I found on the ftp was the 2.6.20.4
1178467364 M * Bertl http://vserver.13thfloor.at/Experimental/patch-2.6.21-vs2.2.0-rc1.diff
1178467383 M * jrsharp the compile error I get is in sched.c, too few args to task_running_tick
1178467385 M * Bertl brcc_: yes, that is true, meminfo is based on rss
1178467398 M * Bertl jrsharp: take the one for 2.6.21 please :)
1178467420 M * Bertl jrsharp: and let us know how it goes ...
1178467435 M * jrsharp yes
1178467437 M * jrsharp definitely
1178467441 M * brcc_ Bertl: Any workaround i can do over there?
1178467517 M * Bertl you mean, to get the supposedly 'wrong' values?
1178467528 M * brcc_ yes
1178467541 M * brcc_ Wrong values but something that can be compared to what comptetiros have
1178467550 M * brcc_ "comptetitors"
1178467552 M * Bertl interesting ...
1178467588 M * brcc_ The problem is: "why can i run tomcat at a 128mb vps on all ISPS but i cant at yours?"
1178467597 M * Bertl well, it should not be too hard to revert the fixes :) but we are actually more interested in getting the accounting as good as possible
1178467688 M * Bertl but before we continue here, I'd prefer to get a comparison with e.g. UML or a real machine ... could you do some test there with, let's say a fixed number of 100 apache processes and see how much memory is actually used? i.e. get the /proc/meminfo before and after start of those processes?
1178467708 M * brcc_ sure
1178467723 M * brcc_ is it ok to do that with vmware?
1178467731 M * Bertl sure
1178467820 M * brcc_ one momment
1178467824 M * brcc_ going to do tht on my local machine
1178467846 M * Bertl make sure to have a compareable kernel
1178467854 M * Bertl i.e. same arch, same kernel version
1178468008 M * brcc_ Hmm, i dont have that
1178468009 M * brcc_ hehe
1178468011 M * brcc_ http://pastebin.ca/474055
1178468048 M * brcc_ This was not on vmware, was on my local machine. 90 apache proccess are just using 10 MB
1178468071 M * brcc_ While on vs2.2 50 apache proccess are going to use all 128MB
1178468081 M * brcc_ Dont they share ram ?
1178468122 M * brcc_ I dont know how this work, but each fork takes the same ammount of memory as the parent ? Dont they share ?
1178468519 M * brcc_ Do you think that downgrading to old stable would be an immediate sollution to my specific problem?
1178468544 M * Bertl fork itself will not share, clone/thread can share
1178468605 M * brcc_ So in the theory if i have 100 apache proccess they should use (100 * (memory of parent))
1178468623 M * Bertl if they are separate processes, yes
1178468743 M * Bertl so your example uses roughly 10MB for the 100 apache processes?
1178468789 M * brcc_ Yes 10 mb
1178468798 M * brcc_ I just tried on openvz and it behaves the same way
1178468829 M * brcc_ I signed up with an openvz provider just to check out how those vps' work.. :)
1178468831 M * Bertl okay, could you do the same on vs2.2 and capture both, the meminfo and the limits for the guest?
1178468844 M * brcc_ sure
1178468853 M * Bertl (keep the guest unlimited)
1178468864 M * brcc_ ok
1178469128 M * brcc_ i am just waiting for an upload to finish on this guest so i can reboot it with unlimited RSS
1178469140 M * Bertl np
1178469218 M * brcc_ 7 mins left
1178469997 T * * http://linux-vserver.org/ | latest stable 2.2.0, 2.0.3-rc2, devel 2.3.0.12, stable+grsec 2.0.2.1, 2.2.0 | util-vserver-0.30.213 | 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'll forget about the minute ;)
1178469997 T * daniel_hozac -
1178470071 M * brcc_ Bertl: http://paste.uni.cc/15252
1178470332 M * Bertl okay, did you capture the limit/accounting values too?
1178470347 M * Bertl if not, what do they show right now?
1178470480 J * hardwire ~bip@rdbck-1922.wasilla.mtaonline.net
1178470641 J * Orange ~debian@71-39-124-91.pool.ukrtel.net
1178470804 M * brcc_ 1 sec
1178470830 M * brcc_ Can i paste here or create anbother pastebin ?
1178470840 M * jrsharp bertl: the patch compiled just fine now
1178470971 M * brcc_ Bertl: http://paste.uni.cc/15253
1178471140 Q * jrsharp Quit: Leaving
1178471254 Q * Orange Quit: Êóäà íè ïëþíü - âåçäå çíàêîìûå ëèöà... ß ÞçÀþ ÑàÌûÉ ÐåÀëÜíÛé ÑêÐèÏò NeOn Script 8.5! CêÀ÷Àé ñ http://script.i-neon.lv :
1178471718 Q * yarihm Quit: Leaving
1178472746 M * Bertl brcc_: so you would be happier to see the anon as used memory, yes?
1178473069 J * er ~sapan@pool-71-168-215-87.cmdnnj.fios.verizon.net
1178473076 M * Bertl wb er!
1178473092 M * er thanks, wow that was quick
1178473118 M * er almost at the same time as the SYN-ACK 
1178473132 M * Bertl well, I'm known to be fast :)
1178473139 M * er and modest
1178473141 M * er :)
1178473157 M * Bertl :)
1178473251 M * bXi they should shoot people who have utf8 characters in their quit messages
1178473265 M * Bertl without notice :)
1178473347 M * bXi especially when their advertising some mirc script
1178475130 M * brcc_ Bertl: I dunno, i liked how it used to be before
1178475147 M * brcc_ it was more likely a "real" server and the accounting virtuozzo does
1178475182 M * brcc_ The new behavior introduces lots of limitations when running a web hosting enviroment for example
1178475197 M * brcc_ If you have, for example, 10 more web accesses than expected you will reach memory limit
1178475212 M * brcc_ in the old approach you could have 100 more without big problems
1178475222 M * brcc_ (this is what happens in openvz and "real" servers)
1178475226 M * Bertl well, I think we can make that a backwards compatible option or so
1178475253 M * Bertl I have to think about it a little more and do some tests I guess
1178475265 M * brcc_ Ok
1178475278 M * Bertl but if you like, there should be a simple hack to go back to the previous behaviour
1178475313 M * brcc_ Great. i would appreaciate that if it is not hard for me (dont know how to play with kernel source :) )
1178475369 M * brcc_ I think that having ANON to be accounted is great, but it still seems not to be correct if we compare this particular apache's fork  case
1178475504 M * brcc_ What if i go back with "old stable" ?
1178475524 M * Bertl kernel/vserver/limit.c edit the vx_vsi_meminfo/vx_vsi_swapinfo
1178475549 M * Bertl to use whatever limit you consider appropriate (for your purpose) instead of RLIMIT_RSS
1178475684 M * brcc_ After changing that, the output of free will be get from "RMAP" for example?
1178475742 M * Bertl yep, that is the function which controls the free/meminfo
1178475766 M * Bertl you can also add some math there, e.g. only account 1/3rd or so
1178475809 M * brcc_ Great. So to have the old behavior i will change that to use RMAP and on /etc/vservers/rlimits ao will have rmap.hard in spite of rss.hard
1178475822 M * brcc_ mv rss.hard rmap.hard
1178477729 J * dreamind ~dreamind@C2107.campino.wh.tu-darmstadt.de
1178479239 J * dothebart ~willi@xdsl-81-173-172-3.netcologne.de
1178479247 Q * tudenbart Read error: Connection reset by peer
1178479518 Q * dreamind Quit: dreamind
1178481090 Q * ntrs_ Ping timeout: 480 seconds
1178481147 Q * toidinamai Ping timeout: 480 seconds
1178481222 J * toidinamai ~frank@i59F7192E.versanet.de
1178482356 Q * bonbons Quit: Leaving
1178482370 M * daniel_hozac Bertl: hmm, doesn't fork() create COW-mappings of the anonymous pages?
1178482761 M * Bertl hmm ...
1178483658 J * ema ~ema@rtfm.galliera.it
1178483704 Q * er Quit: er
1178483973 J * onox ~onox@kalfjeslab.demon.nl
1178484640 P * newz2000 
1178485304 Q * dna Quit: Verlassend
1178486107 Q * bzed Ping timeout: 480 seconds
1178486379 J * bzed ~bzed@dslb-084-059-121-166.pools.arcor-ip.net
1178487202 J * Aiken ~james@121.45.222.137
1178487569 M * Bertl morning Aiken!
1178487591 M * Aiken hi
1178488078 Q * ema Quit: leaving
1178488431 Q * gerrit Ping timeout: 480 seconds
1178489019 Q * sid3windr Ping timeout: 480 seconds
1178489153 J * gerrit ~gerrit@c-67-160-146-170.hsd1.or.comcast.net
1178489428 Q * Johnnie Read error: Connection reset by peer
1178489482 J * sid3windr luser@bastard-operator.from-hell.be
1178490442 Q * fs Ping timeout: 480 seconds
1178490452 J * fs fs@213.178.77.98
1178492017 J * ntrs_ ntrs@68-188-55-120.dhcp.stls.mo.charter.com
1178492307 Q * onox Quit: zzzz
1178493321 N * DoberMann DoberMann[ZZZzzz]
1178494008 J * toidinamai_ ~frank@i59F74F59.versanet.de
1178494135 Q * Aiken Remote host closed the connection
1178494367 Q * toidinamai Ping timeout: 480 seconds
1178494380 Q * Piet Quit: Piet
1178494883 J * Aiken ~james@ppp222-137.lns2.bne1.internode.on.net
1178494939 Q * ruskie Read error: Connection reset by peer
1178495304 J * ruskie ruskie@ruskie.user.oftc.net
1178495432 Q * gerrit Ping timeout: 480 seconds