Dear list,
I've been browsing the XFS archives and the FAQ and found several useful
hints on how to create the filesystem on a RAID5 for beter alignment and
mount options for better performance. I'd like to find out if the values
I found are correct (or maybe too aggressive) for my setup:
The storage is a BR1600 FC2IDE
(http://www.axus.com.tw/raid_br1600fc.htm), the 16 160GB disks are split
in two RAID5 arrays of n=7 disks each and two hot spares. The stripe
size is 128KB. According to Axus support, the data block size is 128
(stripe size) * 512bytes = 64KB.
The server is a dual Xeon 2.8GHz Dell PE4600 with 4GB RAM running RedHat
7.3 (2.4.20-28_36.rh7.3.atsmp). The purpose of the machine is to serve
files over SMB and NFS solely.
For the file system creation, this means:
Data section options:
sunit = 128KB/512bytes = 256
swidth = (n-1)*sunit = 1536
Log section options:
size = 64m
sunit = 64KB/512bytes = 128
The mkfs.xfs command will be:
mkfs.xfs -d sunit=256,swidth=1536 -l size=64m,sunit=128 -L "H1" /dev/...
For the mount options:
logbufs = 16
logbsize = 65536
I understand that increasing those mount values above the default will
improve performance at the cost of more data being lost in case of a
server crash. I'm using the sync option for NFS exports which should
help this issue, though.
Please let me know if you see any problems in my
"calculations/settings". Any advice will be appreciated.
TIA
Regards,
Marc
|