> Justin Gombos wrote: > >> There is a ZFS FAQ somewhere indicating that ZFS *appears* to hog >> memory because it uses as much as it can, but it supposedly >> relinquishes memory as soon as an app calls for it. >> >> I thought that should be pointed out, though I am not quick to accept >> it myself. How does the kernel ask ZFS processes for more memory? > > ZFS uses kmem_cache_create() and friends, which is a fun little beast. > It can manufacture and recycle pre-initialized bits of memory without > having to do much to it. One of the arguments used in setting up a > cache is a callback routine the kernel memory allocator can call when > it needs subsystems to clean up and return memory to the system. > > In the particular case of the ZFS ARC cache, there's a hdr_cache and > a buf_cache; the hdr_cache has a callback, which wakes up the ARC > reclaim thread to evict unneeded allocations in several areas, not > just that one cache.
This has probably been asked before but is there a kernel tunable that will restrict the high water mark for ZFS cache memory? This way I could just grant 50% of mamory and no more. That sort of thing. A more important question would be, what effect would this have on system performance and am I merely thwarting progress. Sort of like early owners of the automobile were forced to send someone walking in front of them with a light in order to ensure that the horses were not frightened. That sort of thing. -- Dennis Clarke _______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org
