P5WD2 Preminum Intel RAID configuration question.

[bogus]

I have four Raptor SATA drives on ports 1-4 and am very pleased with the
performance so far. Currently, I have a RAID10 set using about 95% of
the storage and created a small second RAID0 set with the idea of
placing the Windows Swap file here.

Has anyone done any testing or may have insight into performance issues
with memory intensive applications under this configuration? Will a
swap file located on a RAID0 set provide much of a performace boost? Is
there much of a performance hit having two raid sets on the same 4 drives?

Thx

 

[Paul]

I have four Raptor SATA drives on ports 1-4 and am very pleased with the
performance so far. Currently, I have a RAID10 set using about 95% of
the storage and created a small second RAID0 set with the idea of
placing the Windows Swap file here.

Has anyone done any testing or may have insight into performance issues
with memory intensive applications under this configuration? Will a
swap file located on a RAID0 set provide much of a performace boost? Is
there much of a performance hit having two raid sets on the same 4 drives?

Thx

Backgrounder on the Paging file
http://www.theeldergeek.com/paging_file.htm

Mentions page size is 4KB.
http://support.microsoft.com/default.aspx?scid=kb;en-us;555223

I guess the first question I would ask, is what exactly is the
access pattern of Windows paging ? Does it go to disk each
time that it needs 4KB, or is there some caching or reordering
possible ? If the only primitive it uses is "read 4K" or
"write 4K", I would think a RAID0 stripe wouldn't help at all.
Each operation would tend to only use one of the disks at a
time ?

Paul

 

[Mercury]

I guess the first question I would ask, is what exactly is the

access pattern of Windows paging ?

depends on apps, usage, wind direction, OS.

Does it go to disk each
time that it needs 4KB, or is there some caching or reordering
possible ? If the only primitive it uses is "read 4K" or
"write 4K", I would think a RAID0 stripe wouldn't help at all.
Each operation would tend to only use one of the disks at a
time ?

It supports async / overlapped IO inherently, so more volumes is better.

Paul

Yes, but the IO's would be spread across 4 volumes. You would need a very
busy system that is near / at thrashing (execessive memory use to the point
of system degradation) and so queue IO's to most benefit from the config.
This is most likely on a server system. Getting more RAM would be better in
this situation...

IMHO, *in this case* the page file should be on the first partition on the
system since (depending on defrag program) Windows is likely to be near the
front of the disc(s). Keeping the two adjacent will reduce seeks
particularly on defragged large volumes (OK Raptors aren't that big, but are
fast). This partition should be no bigger than either swap file size or max
swap file size. If you can't do all of this then you are likely to elongate
seek times.

Having the swap file on a RAID 1 (or 10) will not help although from my
experience it doesn't hurt too much (no benchmarks).

Don't forget there are a *lot* of myths about page files. For authentic
info, go to MS. Read up about Performance Monitor and Page Faults.

I suggest you benchmark before and after. Now, what would be a good
benchmark?

 

[bogus]

depends on apps, usage, wind direction, OS.




It supports async / overlapped IO inherently, so more volumes is better.




Yes, but the IO's would be spread across 4 volumes. You would need a very
busy system that is near / at thrashing (execessive memory use to the point
of system degradation) and so queue IO's to most benefit from the config.
This is most likely on a server system. Getting more RAM would be better in
this situation...

IMHO, *in this case* the page file should be on the first partition on the
system since (depending on defrag program) Windows is likely to be near the
front of the disc(s). Keeping the two adjacent will reduce seeks
particularly on defragged large volumes (OK Raptors aren't that big, but are
fast). This partition should be no bigger than either swap file size or max
swap file size. If you can't do all of this then you are likely to elongate
seek times.

Having the swap file on a RAID 1 (or 10) will not help although from my
experience it doesn't hurt too much (no benchmarks).

Don't forget there are a *lot* of myths about page files. For authentic
info, go to MS. Read up about Performance Monitor and Page Faults.

I suggest you benchmark before and after. Now, what would be a good
benchmark?

Interesting points. For my own interest I will investigate this furter,
you have given me some things to think about. I have plenty of RAM at
the moment but I thought setting up with RAID0 swap partition can only
improve the performance and hey, I have no other use for Matrix ATM so
why not use it this way.

 

[Paul]

I suggest you benchmark before and after. Now, what would be a good
benchmark?

On my old Solaris machine at work, I had a 10 line C program that
would malloc all the memory on the machine. The program loops
until it cannot get any more memory. The program just exits and
all the memory is returned to the OS. You get to hear
"lots of grinding".

It is an easy way to test the response of the system, to an
extreme memory load. Using a stopwatch, all you'd need to do,
is measure the time the program remained running. So as not
to bias the results, reboot the computer before each test.

I'm not sure how easy it would be to write that 10 line
program for Windows

Paul

 

[bogus]

I suggest you benchmark before and after. Now, what would be a good
benchmark?

On my old Solaris machine at work, I had a 10 line C program that
would malloc all the memory on the machine. The program loops
until it cannot get any more memory. The program just exits and
all the memory is returned to the OS. You get to hear
"lots of grinding".

It is an easy way to test the response of the system, to an
extreme memory load. Using a stopwatch, all you'd need to do,
is measure the time the program remained running. So as not
to bias the results, reboot the computer before each test.

I'm not sure how easy it would be to write that 10 line
program for Windows

Paul[/QUOTE]

Ya, do not have any projects going right now so I am bored. I have
started writing an application to benchmark the virtual memory
subsystem. I am not sure if it will work out but if it works out maybe
ill post it somewhere. The benchmark program will definitely be
multi-threaded and will obviously need to saturate the physical ram to
be of use.

And one other thought to the suggestion of putting the swap drive at the
beginning of the "matrix" - Windows XP will want to put the OS boot
files on the first "physical" hard drive. This would mean that if the
RAID0 (4 drives in my case) failed then the OS would not boot.

As for the 4k window disk access, I had made the RAID0 use 4k stripes.
I would assume this would make more sense than 128k.

Anyway, I will run the benchmarks and post what I find incase anyone is
interested.