Messages per Month
31 messages in org.freebsd.freebsd-fsThinking of using ZFS/FBSD for a back...| From | Sent On | Attachments |
|---|---|---|
| mike | Jul 8, 2008 6:41 am | |
| CZUCZY Gergely | Jul 8, 2008 8:23 am | |
| mike | Jul 8, 2008 8:31 am | |
| CZUCZY Gergely | Jul 8, 2008 8:34 am | |
| mike | Jul 8, 2008 4:22 pm | |
| CZUCZY Gergely | Jul 8, 2008 7:14 pm | |
| Kris Kennaway | Jul 8, 2008 7:49 pm | |
| CZUCZY Gergely | Jul 8, 2008 8:13 pm | |
| Kris Kennaway | Jul 8, 2008 8:34 pm | |
| Ollivier Robert | Jul 8, 2008 8:40 pm | |
| Ollivier Robert | Jul 8, 2008 8:42 pm | |
| CZUCZY Gergely | Jul 8, 2008 8:54 pm | |
| Kris Kennaway | Jul 8, 2008 9:09 pm | |
| Randy Bush | Jul 8, 2008 9:47 pm | |
| Randy Bush | Jul 8, 2008 9:48 pm | |
| mike | Jul 8, 2008 10:31 pm | |
| Kris Kennaway | Jul 8, 2008 10:40 pm | |
| Kris Kennaway | Jul 8, 2008 10:40 pm | |
| Randy Bush | Jul 8, 2008 10:47 pm | |
| Kris Kennaway | Jul 8, 2008 11:14 pm | |
| Kris Kennaway | Jul 8, 2008 11:21 pm | |
| Dmitry Morozovsky | Jul 9, 2008 2:36 am | |
| CZUCZY Gergely | Jul 9, 2008 5:44 am | |
| Jeremy Chadwick | Jul 9, 2008 5:56 am | |
| Kris Kennaway | Jul 9, 2008 10:18 am | |
| Kris Kennaway | Jul 9, 2008 10:25 am | |
| CZUCZY Gergely | Jul 9, 2008 10:37 am | |
| Dmitry Morozovsky | Jul 9, 2008 10:52 am | |
| Kris Kennaway | Jul 9, 2008 11:16 am | |
| Randy Bush | Jul 9, 2008 11:50 am | |
| Lorenzo Perone | Jul 9, 2008 1:34 pm |
| Subject: | Thinking of using ZFS/FBSD for a backup system | |
|---|---|---|
| From: | Kris Kennaway (kr...@FreeBSD.org) | |
| Date: | Jul 8, 2008 9:09:54 pm | |
| List: | org.freebsd.freebsd-fs | |
Ollivier Robert wrote:
According to Kris Kennaway:
(ports-i386:~)> cat /boot/loader.conf vfs.zfs.prefetch_disable=1 vm.kmem_size=1572864000
Hvae you tried w/o the prefetch_disable tunable?
I have not done careful measurements, but casual observation suggests that on my workloads I get better performance without prefetch. I have a pair of mirrored disks on an amr, and my workload is quite random-access so prefetching just introduces latencies and wastes already-saturated disk bandwidth. With more disks or a different workload I would expect different performance characteristics.
Kris