Snapshot duration, performance and how to avoid I/O lock
Ulrich Spörlein
ulrich.spoerlein at 1822direkt.com
Wed Sep 6 06:36:21 PDT 2006
Hi,
I have to create regular snapshots of several volumes roughly 1.4TB in
size (each). But using mksnap_ffs takes a lot of time (45 minutes) and
it looks like it could be speed up.
iostat reports some 2MB/s of I/O
tty da0 da1 sa0
cpu
tin tout KB/t tps MB/s KB/t tps MB/s KB/t tps MB/s us ni sy in id
0 77 16.00 120 1.87 0.00 0 0.00 0.00 0 0.00 0 0 8 4 89
0 231 16.00 131 2.04 0.00 0 0.00 0.00 0 0.00 0 0 6 2 91
0 77 16.00 127 1.98 0.00 0 0.00 0.00 0 0.00 0 0 5 3 92
0 204 16.00 123 1.92 0.00 0 0.00 0.00 0 0.00 0 0 6 3 91
0 77 16.00 128 2.00 0.00 0 0.00 0.00 0 0.00 0 0 6 3 91
gstat reports
dT: 0.501s w: 0.500s
L(q) ops/s r/s kBps ms/r w/s kBps ms/w %busy Name
1 126 84 1341 11.8 42 671 2.7 99.7| da0
The filesystem under snapshotting was *empty*, and right now is at
Filesystem 1K-blocks Used Avail Capacity Mounted on
/dev/da0s2d 1376038286 815092 1265140132 0% /export/vol1
How come, a snapshot of an empty file system takes more than 800MB? Aren't the
blocks copy-on-write?
Why is the disk 100% busy while only a mere 2MB/s are pushed? This is on a
3ware device driver for 9000 series storage controllers, version: 3.60.02.012
twa0: <3ware 9000 series Storage Controller> port 0x3000-0x303f mem 0xdc000000-0xddffffff,0xd8300000-0xd8300fff irq 48 at device 1.0 on pci3
twa0: [FAST]
twa0: INFO: (0x15: 0x1300): Controller details:: Model 9550SX-8LP, 8 ports, Firmware FE9X 3.04.00.005, BIOS BE9X 3.04.00.002
with a RAID5 over 7 SATA disks.
da0 at twa0 bus 0 target 0 lun 0
da0: <AMCC 9550SX-8LP DISK 3.04> Fixed Direct Access SCSI-3 device
da0: 100.000MB/s transfers
da0: 1430448MB (2929557504 512 byte sectors: 255H 63S/T 182356C)
Another thing is blocking other disk I/O while snapshotting. Right now I did
a ls(1) in the .snap directory, so I understand the filesystem is now suspended.
The workaround would then be to "dont do that". But what if other snapshots are
accessed during that time? I want to provide yesterdays snapshot to our users
while taking the current snapshot and providing access to the newest data at the
same time.
Cheers
Ulrich Spoerlein
More information about the freebsd-stable
mailing list