[zones-discuss] rcapd
Hi , I am facing an issue with rcapd, currently I have setup 8 sparse-root containers on a server with 32G physical memory , I have capped each of these containers varyingly and there is no issue with capping and it works fine. The issue arises when one of the containers eats up more memory (rapidly) than it has been allocated .It causes other non global zones to be less (noticable ) responsive while rcapd is trying to curb this unruly behaviour by one of the containers.I am wondering if this is due to heavy paging ? Has anyone else seen such behaviour, or is this an acceptable behaviour ? Any comments or experiences would be really helpful . Thanks -- This message posted from opensolaris.org ___ zones-discuss mailing list zones-discuss@opensolaris.org
Re: [zones-discuss] rcapd
Hi Syed, I would not be surprised to find that rcapd is behaving correctly on your system. All of the containers in one Solaris instance share one Solaris paging system and one set of swap devices. When rcapd is paging the memory pages of one container out to the swap device, other workloads sharing that disk will take longer to write to that disk. This is similar to other virtualized solutions (e.g. hypervisors) that have similar constraints, similar workloads and are sharing one internal disk for swap space. If your other containers are not paging at all, you can reduce this effect by configuring your swap space on its own disk drive. The disk-write transactions from those other containers will then *not* wait for paging activity of the container with a RAM cap that is too low. Do you know why that one container is using up more memory than the cap? Is the cap too low, or the application behaving badly? On Mon, Sep 1, 2008 at 7:55 AM, syed [EMAIL PROTECTED] wrote: Hi , I am facing an issue with rcapd, currently I have setup 8 sparse-root containers on a server with 32G physical memory , I have capped each of these containers varyingly and there is no issue with capping and it works fine. The issue arises when one of the containers eats up more memory (rapidly) than it has been allocated .It causes other non global zones to be less (noticable ) responsive while rcapd is trying to curb this unruly behaviour by one of the containers.I am wondering if this is due to heavy paging ? Has anyone else seen such behaviour, or is this an acceptable behaviour ? Any comments or experiences would be really helpful . -- --JeffV ___ zones-discuss mailing list zones-discuss@opensolaris.org
Re: [zones-discuss] rcapd
On Mon, Sep 1, 2008 at 6:55 AM, syed [EMAIL PROTECTED] wrote: Hi , I am facing an issue with rcapd, currently I have setup 8 sparse-root containers on a server with 32G physical memory, I have capped each of these containers varyingly and there is no issue with capping and it works fine. The issue arises when one of the containers eats up more memory (rapidly) than it has been allocated .It causes other non global zones to be less (noticable ) responsive while rcapd is trying to curb this unruly behaviour by one of the containers.I am wondering if this is due to heavy paging ? What does vmstat -p say? I bet it says yes! Has anyone else seen such behaviour, or is this an acceptable behaviour ? Any comments or experiences would be really helpful. I haven't, but then again that is because I expected to see such behavior if I used rcapd. There are very few circumstances in my world that make sense to encourage heavy paging - as rcapd will do. Solid state disk may change this a bit because paging would likely be a lot faster. For now, my approach is to cap the use of swap. Note that in this definition, swap is different than most people expect - it refers to how much memory a zone can reserve. If the sum of all of your zones' swap cap is 32 GB, you should see pretty much no paging to swap devices. You will still see file system (e.g. executable) paging. The most noticable side effect is that if the things running in a zone (including use of /tmp) try to use more than their respective caps, memory allocations will fail. I see this as a good thing because it means that the misbehaving application fails rather than taking down all the rest. rcapd works a bit like boot camp (military - not the mac thing). If one soldier (zone) misbehaves they all get punished. There may be circumstances where this outcome is desirable but server virtualization using zones is likely not one of them. -- Mike Gerdts http://mgerdts.blogspot.com/ ___ zones-discuss mailing list zones-discuss@opensolaris.org
Re: [zones-discuss] rcapd
Hi Mike, I think you are right, using rcapd for virtulization using zones would be a problem, thanks for the work around , really appreciate your help. Regards uzair --- On Mon, 9/1/08, Mike Gerdts [EMAIL PROTECTED] wrote: From: Mike Gerdts [EMAIL PROTECTED] Subject: Re: [zones-discuss] rcapd To: syed [EMAIL PROTECTED] Cc: zones-discuss@opensolaris.org Date: Monday, September 1, 2008, 9:24 AM On Mon, Sep 1, 2008 at 6:55 AM, syed [EMAIL PROTECTED] wrote: Hi , I am facing an issue with rcapd, currently I have setup 8 sparse-root containers on a server with 32G physical memory, I have capped each of these containers varyingly and there is no issue with capping and it works fine. The issue arises when one of the containers eats up more memory (rapidly) than it has been allocated .It causes other non global zones to be less (noticable ) responsive while rcapd is trying to curb this unruly behaviour by one of the containers.I am wondering if this is due to heavy paging ? What does vmstat -p say? I bet it says yes! Has anyone else seen such behaviour, or is this an acceptable behaviour ? Any comments or experiences would be really helpful. I haven't, but then again that is because I expected to see such behavior if I used rcapd. There are very few circumstances in my world that make sense to encourage heavy paging - as rcapd will do. Solid state disk may change this a bit because paging would likely be a lot faster. For now, my approach is to cap the use of swap. Note that in this definition, swap is different than most people expect - it refers to how much memory a zone can reserve. If the sum of all of your zones' swap cap is 32 GB, you should see pretty much no paging to swap devices. You will still see file system (e.g. executable) paging. The most noticable side effect is that if the things running in a zone (including use of /tmp) try to use more than their respective caps, memory allocations will fail. I see this as a good thing because it means that the misbehaving application fails rather than taking down all the rest. rcapd works a bit like boot camp (military - not the mac thing). If one soldier (zone) misbehaves they all get punished. There may be circumstances where this outcome is desirable but server virtualization using zones is likely not one of them. -- Mike Gerdts http://mgerdts.blogspot.com/ ___ zones-discuss mailing list zones-discuss@opensolaris.org
[zones-discuss] rcapd interactions between global and local zones?
I'm struggling with the resource capping documentation. This environment uses zones and the documentation is frustratingly vague about resource capping and zones. A) If I run rcapd in the global zone and have resource settings via zonecfg for one or more zones, it would seem to me that I would not want to run rcapd in any local zones with capped-memory settings since that since both might end up working on the same process and possibly deadlock? B) If I do not have capped-memory settings for the zones in the zonecfg, I should be able to run rcapd in the global zone and rcapd in local zones? Any other limitations? Jim Litchfield ___ zones-discuss mailing list zones-discuss@opensolaris.org
Re: [zones-discuss] Rcapd threshold vs ZFS cache
Brian Smith wrote: When rcapd is calculating how much memory is free, to compare to the memory cap enforcement threshold, does it consider the memory used by the ZFS cache to be free or used? If I set rcapadm -c 90 then will rcapd behave any differently than with rcapadm -c 0 if there is a ZFS cache? I have read that the ZFS cache will use all free physical RAM. IIRC, the ARC is limit to 3/4 of total memory. That is the max, it may use [much] less, depending on overall memory use. In your case, where you are allowing up to 90%, the ARC may get small if that memory is actually getting used. zfs-discuss is a good place to get info regarding ARC vs VM memory Regards, Brian ___ zones-discuss mailing list zones-discuss@opensolaris.org ___ zones-discuss mailing list zones-discuss@opensolaris.org
Re: [zones-discuss] rcapd to limit RAM usage of a zone
Phil Cordier wrote: Posted this question on the general zones group at got a deafening silence in response - anyone here have any possible answers? http://forum.sun.com/jive/thread.jspa?forumID=299threadID=100707 It is possible that prstat and rcapd are counting shared memory multiple times, so that it seems like you are using more memory within the zone. This is a known bug: 4754856 *prstat* prstat -atJTZ should count shared segments only once We are working on an enhancement to zones and resource management which will improve a bunch of things in this area, including the way rcapd accounts for shared memory. It will also allow you to solve your underlying problem by running a single rcapd in the global zone which will cap each zones memory consumption. This is described in this thread: http://www.opensolaris.org/jive/thread.jspa?threadID=10451tstart=0 Let us know if you have any questions about this project. Thanks, Jerry ___ zones-discuss mailing list zones-discuss@opensolaris.org
Re: [zones-discuss] rcapd to limit RAM usage of a zone
Funny but ironic: rcapd never counts its own memory usage. See the source code at http://cvs.opensolaris.org/source/search?q=rcapd_pid . Once you subtract rcapd's 128MB, the *rest* of the zone is using exactly what you specified as the cap. But I don't see why rcapd should be using 128MB... Can you supply the output of zone11# pmap -ax {pid-of-rcapd} Phil Cordier wrote: Posted this question on the general zones group at got a deafening silence in response - anyone here have any possible answers? http://forum.sun.com/jive/thread.jspa?forumID=299threadID=100707 -- -- Jeff VICTOR Sun Microsystemsjeff.victor @ sun.com OS AmbassadorSr. Technical Specialist Solaris 10 Zones FAQ:http://www.opensolaris.org/os/community/zones/faq -- ___ zones-discuss mailing list zones-discuss@opensolaris.org