RE Tuning for performance
2006-08-21 11:36:23
Hi all,
I would like to proceed with my thread about server
tuning for BOT performance.
The goal is to summarize in one place what I did,
I'm doing, and will do, to improve my BOT system performance in the hope
that it will helps others one day.
Tape drives are becoming faster than ever (even faster
than HD) and I think, in the future, servers will have to be more and more
optimized to keep up with this ever-increasing rate.
For the record my BOT system is made of:
1) 1 Dell Poweredge 2850 with 1 1.4TB
RAID5 array made of 6 U320 SCSI HD
2) 1 Dell Powervault 110 LTO3 tape drive
3) Redhat ES 3.3
4) Amanda (amanda-2.4.4p1-0.3E)
As I only have 1 logical disk, I decided to not use
the "holding disk" feature of Amanda thinking that it could only
lead to worse performance (Amanda would have to read the data from the
disk, write an archive from it, and then finally read it again to write
it on tape. All these IO being made on the same disk. I believe that it
is slower than having Amanda writing directly to tape)
Initial situation:
==================
The BOT speed is about 20 MB/s. Most probably due
to HD bottleneck.
Tuning performed till now:
==========================
1) raised the "tapebufs" parameter in Amanda
to 1024.
Didn't helped. The only difference I
see is this little message in the "NOTE" section from time to
time (which I believe doesn't harm):
NOTES:
taper: attach_buffers: (1024 tapebufs:
33570816 bytes) Invalid argument
2) Mount the filesystem with the "noatime"
flag
Didn't helped.
3) Modify the SCSI RAID card configuration: Changed
the "read policy" from "adaptive" to "readahead"
and the "cache policy" from "directIO" to "cachedIO"
Didn't helped.
4) raised the "/proc/sys/vm/min-readahead"
and "/proc/sys/vm/max-readahead" parameters from 3 and 31 to
33 and 255.
Helped a bit. Mostly for full backups.
Yesterday, I could achieve an average
dump rate of 40 MB/s (42 MB/s for full and 14 MB/s for daily).
STATISTICS:
Total
Full Daily
-------- --------
--------
Estimate Time (hrs:min) 1:03
Run Time (hrs:min)
2:16
Dump Time (hrs:min)
1:13 1:08 0:04
Output Size (meg) 173951.1
170245.2 3706.0
Original Size (meg) 173951.1
170245.2 3706.0
Avg Compressed Size (%)
-- -- --
(level:#disks ...)
Filesystems Dumped
28 19
9 (1:9)
Avg Dump Rate (k/s) 40866.5
42578.5 14354.4
What's left to try:
===================
1) Continue to raise the "/proc/sys/vm/min-readahead"
and "/proc/sys/vm/max-readahead" parameters.
2) Tune the elevator algorithm (elvtune)
3) Upgrade to RH ES 4.3 (new megaraid driver)
4) Split the logical raid drive in 2 (1 for
the OS, 1 for the data)
5) Use the holding disk feature.
6) Recompile Amanda and raise the tape record
size
7) increase the server's TCP/IP MTU?
8) Any other idea?
|
<Prev in Thread] |
Current Thread |
[Next in Thread>
|
- RE Tuning for performance,
Cyrille Bollu <=
|
|
|