UNIX for Dummies Questions & Answers

Hi guys,
what is the relation between I/O performance and file systems.
I have a file systems called /dcs/data01 which is having 4Tb size.
According our application we can split the file system like

dcs/data01 -> 1Tb
dcs/data02 -> 1Tb
dcs/data03 -> 1Tb
dcs/data04 -> 1Tb

do you think I/O performance will increase by doing this.

Many Thanks
Davinzy

That depends on how the data is read, is it sequential or random access, is one process reading the data or many?

The performance is more likely to be affected by the volumes physical make up than the filesystem, i.e. is it striped across several disks, is it mirrored or just a RAID 5? The filesystem would have a more significant impact on performance with many small files but with only 1 or 4 files it will have little effect.

What will improve by splitting the 4TB files into 4 1TB files is the chances of successfully backing up and restoring the data, especially if a restore of only 1 of the 4 files is required!

The best answer is to measure your current performance and compare that with trying 4 separate files, not the answer you wanted I'm sure...

HI Tony,

thank you very much for your reply, we are using ISAM to read data and there are many processors reading data.
Our system stores thousands of files in that file system which are having millions of records. And out processors read and write data frequently to that location.

no backing up of files needed, system will automatically purge files after file retention period .

how can I check how many physical volumes are there ?

The next question is what connection is used to your disks...
FC to a SAN? what speed?
There are may things you can do to improve but for that you need to know your configuration:
SAN?
VG with ONlineJFS or other?
are the disks in the VG from a same array? (SAN side that is...)
Are you using aggregated LUNS?
What parameters are set on the filesystem?
etc...

From what I remember you run:
# df -k .
in the data volume to determine the logical volume (LV) it is stored in.

# lvdisplay -v lvnn
where lvnn is the name and number of the LV your data is stored in to see what volume group your LV is part of.

# vgdisplay vgnn
where vgnn is the name and number of the volume group (VG).

This will tel you how many Physical Volumes (PVs) your VG is spread over, they could be slices of disks some on the same disk just top confuse matters and the VGs could be divided into several LVs but hopefully your system is using whole disks and the VG is dedicated to the LV your data is stored in!

I hope that helps?


Edit:
I should have said that by the sounds of what you say you do need the volume striped over as many disks as possible to get the best bandwidth and access time for your parallel processing on multiple small files.