Date: Mon, 10 Dec 2007 12:16:00 +0200 From: Artis Caune <Artis.Caune@sigmanet.lv> To: fs@FreeBSD.org Subject: +12K ZFS file systems or snapshots Message-ID: <475D11E0.1050701@sigmanet.lv>
next in thread | raw e-mail | index | archive | help
Hello FreeBSD-fs,
We want to make ZFS file system for each user:
# zfs create tank/user_x
# zfs set quota=1G tank/user_x
We also want to create snapshots for every user:
user_x@month_ago
user_x@week_ago
user_x@day_ago
We have run bonnie tests on amd64, 6G RAM IBM box with 3 SAS drives in
raidz1 (7.0-BETA3).
Tests failed with "kmem_map too small".
Raising kmem_size to 1500M and limiting arc_max to 512M passed bonnie
test for 4 days.
'vmstat -m' solaris MemUse is ~600M - 700M
We could not create more than 12K file systems with 'zfs create'.
ZFS panics with "kmem_map too small" when 'vmstat -m' solaris MemUse
reaches 1500M (kmem_size).
Solaris recommend 1G for every 10K file systems or snapshots, each fs or
snapshot requires 64K of kmem.
It looks like kmem_size is u_int and is limited to 4G, but we can boot
only with 1500M. With 2G kmem_size console is full with panics.
We want +/- 10K users on each box. We can:
* create 10K file systems (users), use tank snapshots, not per-user
snapshots
* create 3K file systems (users) and 3 snapshots per file system
+room for zfs cache
If kmem_size is limited to ~2G, then there is no need for more than 2G
RAM for ZFS file server?
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?475D11E0.1050701>