Greetings, I've setup a new ubuntu 10.10 server for xen 4.0.1, though it included kernel compilation (creating PVOPS 2.6.32 kernel) and all the stuff by their guide - (generally include xen and download all packages needed and make world + add sata drivers into kernel).
Now everything is running fine, though I've noticed that dom0 takes in too much memory for nothing - right after dom0 is started - 860 MB is marked as used, even with all caches dropped and only service is SSH and XEND (I can turn even this one off and memory won't free)... and it's apparently not some XEN balloon-memory thing- when I "xm mem-set 0 800" it runs OOM-killer and end in restart ....
meminfo is confusing me at this :
cat /proc/meminfo MemTotal: 1229376 kB MemFree: 345440 kB Buffers: 412 kB Cached: 8572 kB SwapCached: 0 kB Active: 45304 kB Inactive: 14112 kB Active(anon): 38464 kB Inactive(anon): 12304 kB Active(file): 6840 kB Inactive(file): 1808 kB Unevictable: 20 kB Mlocked: 20 kB SwapTotal: 0 kB SwapFree: 0 kB Dirty: 948 kB Writeback: 0 kB AnonPages: 50508 kB Mapped: 6600 kB Shmem: 320 kB Slab: 15904 kB SReclaimable: 5236 kB SUnreclaim: 10668 kB KernelStack: 1016 kB PageTables: 1664 kB NFS_Unstable: 0 kB Bounce: 0 kB WritebackTmp: 0 kB CommitLimit: 614688 kB Committed_AS: 153440 kB VmallocTotal: 34359738367 kB VmallocUsed: 264292 kB VmallocChunk: 34359473836 kB HugePages_Total: 0 HugePages_Free: 0 HugePages_Rsvd: 0 HugePages_Surp: 0 Hugepagesize: 2048 kB DirectMap4k: 8647280 kB DirectMap2M: 0 kB
last lines stand out in particular: DirectMap4k: 8647280 kB what is this direct map ? Apparently it's size I cannot go under, but neither I can use for userspace... anybody could give me a hint what direct map is ? and how can I make it smaller perhaps if that makes any sense. Machine CPU is AMD Opteron 6168.
Could this be caused by some kernel misconfiguration?
By comparison, other xen servers eat just around 100-200 MB for dom0 and their vital processess. I could always clone xen-3.3 image from them, but I've wanted new xen to benefit from the new smp-impoved blk devices ...
just run
top
and sort by memoryM
. See which, if any process, is consuming memory. I found in older versions on one of my servers that one of the xen python scripts was slowly eating more memory the more I ranxm list
every time. It seemed to have happened to one host which since has moved to Debian 6.0 and no issues anymore. Your case could be related some other process or something else.