C
C. Morrow
I am planning on a 4disk RAID 5 for my desktop PC. I have used the on-board
Raid 0 implementation on my Soyo Dragon+ mobo and loved the increase in
performance I got BUT I got bit when one of the drives failed and I lost
everything. (I had #$%^& IBM 75 and 60 GXP drives).
I am looking for performance increase, both reads and writes as well as data
security so now onto RAID 5. Security is a given with RAID 5 but my
question has to do with performance, in particular performance for a desktop
PC, not a server.
In researching which IDE RAID 5 controller to get, I discovered there are
hardware and software implementations. The main difference being that
hardware implementations have a coprocessor to do the checkdigit (XOR)
calculation while the software RAID uses the processor's CPU to do the
calculations.
So I would expect the software RAID to have higher CPU utilization compared
to hardware RAID. So far so good and it would seem that the harware
solution should have better performace.
BUT since this is for a desktop PC, not a server, I am not so sure that the
software RAID would have worse performance than hardware RAID. In fact, my
processor (AMD XP2100) can probably do the checkdigit calculations FASTER
than the onboard processor on a hardware RAID board. The closest I've come
to verifying this is in the review I read at tech-report.com
http://www.tech-report.com/reviews/2002q4/ideraid/index.x?pg=1 where the
Adaptec 2400A hardware RAID board came out with the worst performance, even
when compared to the software RAID Promise SX4000. In fact the Promise
SX4000 had the best RAID 5 performance in many areas.
Now I can understand in the server world, it is best to offload the RAID 5
checksum calculations so the server processor ca go and do other things.
But in the desktop PC world, where I am the only user and when I do a disk
operation, I am waiting for it to complete anyway, I do not think that
having my processor do some of the work should affect the performance that I
see. Especially since there would be few other tasks going on anyway.
So the conclusion I come to is that for desktop PC, software RAID 5 should
be an acceptable performance solution. And it certainly is cheaper.
What do you think? If you disagree, can you point me to some references
that proves the point?
cym
Raid 0 implementation on my Soyo Dragon+ mobo and loved the increase in
performance I got BUT I got bit when one of the drives failed and I lost
everything. (I had #$%^& IBM 75 and 60 GXP drives).
I am looking for performance increase, both reads and writes as well as data
security so now onto RAID 5. Security is a given with RAID 5 but my
question has to do with performance, in particular performance for a desktop
PC, not a server.
In researching which IDE RAID 5 controller to get, I discovered there are
hardware and software implementations. The main difference being that
hardware implementations have a coprocessor to do the checkdigit (XOR)
calculation while the software RAID uses the processor's CPU to do the
calculations.
So I would expect the software RAID to have higher CPU utilization compared
to hardware RAID. So far so good and it would seem that the harware
solution should have better performace.
BUT since this is for a desktop PC, not a server, I am not so sure that the
software RAID would have worse performance than hardware RAID. In fact, my
processor (AMD XP2100) can probably do the checkdigit calculations FASTER
than the onboard processor on a hardware RAID board. The closest I've come
to verifying this is in the review I read at tech-report.com
http://www.tech-report.com/reviews/2002q4/ideraid/index.x?pg=1 where the
Adaptec 2400A hardware RAID board came out with the worst performance, even
when compared to the software RAID Promise SX4000. In fact the Promise
SX4000 had the best RAID 5 performance in many areas.
Now I can understand in the server world, it is best to offload the RAID 5
checksum calculations so the server processor ca go and do other things.
But in the desktop PC world, where I am the only user and when I do a disk
operation, I am waiting for it to complete anyway, I do not think that
having my processor do some of the work should affect the performance that I
see. Especially since there would be few other tasks going on anyway.
So the conclusion I come to is that for desktop PC, software RAID 5 should
be an acceptable performance solution. And it certainly is cheaper.
What do you think? If you disagree, can you point me to some references
that proves the point?
cym