If I understand it correctly, the problem in this case is that even when you do have more than enough RAM, kernel just does not provide enough space in kmem_map to map that physical memory into. Unless you want to have dedupe turned on large filesystem (and FreeBSD does not have this feature yet), you don't really *need* all that much RAM. "kmem_map too small" comes up as an issue way more often than lack of physical memory. --Artem On Thu, Apr 29, 2010 at 9:56 AM, Scott Long <scottl_at_samsco.org> wrote: > On Apr 29, 2010, at 9:44 AM, Tom Evans wrote: >> On Thu, Apr 29, 2010 at 3:53 PM, Ollivier Robert >> <roberto_at_keltia.freenix.fr> wrote: >>> According to James R. Van Artsdalen: >>>> system is a Core i7 975 (3.33 GHz x 4 cores 3x threads per core) with 12 >>>> GB of RAM, a 2x2TB ZFS boot pool and a second (idle) pool of 16x2TB. >>> >>>> panic: kmem_malloc(131072): kmem_map too small: 3832475648 total allocated >>> >>> Apart from the fact that you must at least set vm.kmem_size to something like 2x your RAM, one rule of thumb I've seen discussed for ZFS is that you will need approximatively 1 GB of RAM per TB of data so you may be a bit short here to get optimal perfs. >>> >> >> Citation needed? I have a file server running amd64 8-STABLE with 4GB >> of RAM, 6 x 1.5 TB drives in raidz, and have never had any problems >> with memory usage. Are you saying that after my next update, adding >> another 6 x 1.5 TB drives, it will start being flaky and/or panicing >> with kmem_map too small errors? >> > > I'm sorry, but I find it absolutely absurd that any filesystem has to wire down 2GB of RAM, and that the solution to panics is buy more RAM. > > Scott > > _______________________________________________ > freebsd-current_at_freebsd.org mailing list > http://lists.freebsd.org/mailman/listinfo/freebsd-current > To unsubscribe, send any mail to "freebsd-current-unsubscribe_at_freebsd.org" >Received on Thu Apr 29 2010 - 16:25:55 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:03 UTC