For whatever it is worth, if I use geli rather than gbde I get normal
(~30MB/s) performance.
I also get the same slow gbde performance on 8.1-PRERELEASE as of last night.
I've make a kernel swaping in files from 7.2 source to see if I got
any improvement.
I pulled in:
geom_dev.c (with some hacks to get it compile)
geom_slice.c
geom_io.c
None of those improved performance.
If anyone has any suggestions for things to try, let me know. I am
fine with switching to geli for some applications but I have about
12TB under gbde. That would be somewhat of a bear to copy over into
geli.
On Mon, May 17, 2010 at 7:32 PM, Joseph Gleason fired...@gmail.com wrote:
Sometime between FreeBSD 7.2-RELEASE-p4 and 8.0-RELEASE write
performance of gbde encrypted devices seems to have dropped
significantly. A system I have running 7.2 seems to run gbde drives
at or near the drive max rate (30-40MB/s) while I am seeing less than
10% of that on 8.0 systems.
I get the same slow writes on 8.0-RELEASE-p2 as well as 8.0-RELEASE.
Here is an example on a fresh 8.0 install which shows gbde taking the
drive write performance of 40 MB/s down to 2.6 MB/s:
lab# uname -a
FreeBSD lab.int.fireduck.com 8.0-RELEASE FreeBSD 8.0-RELEASE #0: Sat
Nov 21 15:02:08 UTC 2009
r...@mason.cse.buffalo.edu:/usr/obj/usr/src/sys/GENERIC amd64
lab# dd if=/dev/urandom of=/dev/ad4s1d bs=32k count=32k
32768+0 records in
32768+0 records out
1073741824 bytes transferred in 25.130537 secs (42726577 bytes/sec)
lab# gbde init /dev/ad4s1d
Enter new passphrase:
Reenter new passphrase:
lab# gbde attach /dev/ad4s1d
Enter passphrase:
lab# dd if=/dev/urandom of=/dev/ad4s1d.bde bs=32k count=32k
32768+0 records in
32768+0 records out
1073741824 bytes transferred in 401.097004 secs (2677013 bytes/sec)
iostat from while that last 'dd' was running:
tty ad4 cpu
tin tout KB/t tps MB/s us ni sy in id
0 22 5.67 483 2.67 0 0 4 1 96
0 66 5.67 509 2.82 0 0 4 1 95
0 22 5.69 514 2.86 0 0 6 1 94
0 22 5.67 506 2.80 0 0 6 1 93
0 22 5.67 472 2.61 0 0 4 1 95
iostat on a FreeBSD 7.2-RELEASE-p4 box doing a similar operation:
tin tout KB/t tps MB/s us ni sy in id
0 22 29.54 1208 34.86 3 0 56 2 39
0 22 29.56 1177 33.97 3 0 57 1 39
0 22 29.54 1201 34.64 3 0 58 2 37
0 22 29.57 1144 33.04 2 0 51 3 44
0 22 29.56 1126 32.52 3 0 54 2 42
0 22 29.53 1179 34.01 3 0 53 2 42
0 22 29.57 1165 33.65 2 0 58 2 38
One thing I notice is the larger block size the 7.2 writes but I don't
imagine that would be that significant.
I've been using FreeBSD in various amateurish and wrong ways since
2.2, so I wouldn't rule out me doing something stupid. If so, I'd
love to know what.
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
