Fastest: 4 Disk RAID 5 or 4 DISK RAID 10?

[Roland Wooster]

So, in XP, using a Promise Serial ATA S140 SX4 (with 256MB cache) and
four disks, 160GB 8GB, 7200rpm, either the Seagate or Samsung (looking
for quiet) which will I get the best performance from: either the 4
disks in a single RAID5 array, or the four disks in a RAID 10 array.
[Yes I know I'm talking 480GB versus 320GB of usable space in the end,
but performance is the question here]

Note that this is likely to be the only disk array in the system, so
would have the OS (WinXP) video editing, photoshop image editing etc
all on the same array. The disk system is most intensively used in
video and image editing.

Alternatively of course I could put 3 disks in the RAID 5 array, and
one disk on it's own as a swap/temp drive if that would be faster for
say the OS swap file, \temp folder, and Photoshop's temp directory
too.

What are your thoughts?

Thanks,
Roland.

 

[Will Dormann]

So, in XP, using a Promise Serial ATA S140 SX4 (with 256MB cache) and
four disks, 160GB 8GB, 7200rpm, either the Seagate or Samsung (looking
for quiet) which will I get the best performance from: either the 4
disks in a single RAID5 array, or the four disks in a RAID 10 array.
[Yes I know I'm talking 480GB versus 320GB of usable space in the end,
but performance is the question here]

RAID 10 will give you higher performance than RAID 5.

-WD

 

[Marc de Vries]

So, in XP, using a Promise Serial ATA S140 SX4 (with 256MB cache) and
four disks, 160GB 8GB, 7200rpm, either the Seagate or Samsung (looking
for quiet) which will I get the best performance from: either the 4
disks in a single RAID5 array, or the four disks in a RAID 10 array.
[Yes I know I'm talking 480GB versus 320GB of usable space in the end,
but performance is the question here]

Note that this is likely to be the only disk array in the system, so
would have the OS (WinXP) video editing, photoshop image editing etc
all on the same array. The disk system is most intensively used in
video and image editing.

That's easy. RAID10 is much faster.
Especially since you will also be writing to the disk.

RAID5 only performs well when sequentially reading large files, but is
slow when writing files.

Alternatively of course I could put 3 disks in the RAID 5 array, and
one disk on it's own as a swap/temp drive if that would be faster for
say the OS swap file, \temp folder, and Photoshop's temp directory
too.

The swapfile isn't used anyway, so there is no gain from putting it on
another drive.
(If your swapfile is used a lot you should buy more memory instead of
RAID)

And what programs could be using a temp drive? The video editing
software is the reason why you want a raid array, so that stays on the
array.

But do you actually need all that disk performance?
Most reasonably intelligent editing programs don't need all that much
diskperformance

Marc

 

[Shailesh Humbad]

Check out my review of that controller, especially since you'll be
running XP:

http://www.somacon.com/fasttrak_sx4/

For video editing, you definitely want RAID-10 over RAID-5. RAID-5 is
slower than a single disk for writing operations, and is really
intended to increase the speed of random-read operations in database
servers.

 

[Roland Wooster]

So, in XP, using a Promise Serial ATA S140 SX4 (with 256MB cache) and
four disks, 160GB 8GB, 7200rpm, either the Seagate or Samsung (looking
for quiet) which will I get the best performance from: either the 4
disks in a single RAID5 array, or the four disks in a RAID 10 array.
[Yes I know I'm talking 480GB versus 320GB of usable space in the end,
but performance is the question here]

Note that this is likely to be the only disk array in the system, so
would have the OS (WinXP) video editing, photoshop image editing etc
all on the same array. The disk system is most intensively used in
video and image editing.

That's easy. RAID10 is much faster.
Especially since you will also be writing to the disk.

RAID5 only performs well when sequentially reading large files, but is
slow when writing files.

Alternatively of course I could put 3 disks in the RAID 5 array, and
one disk on it's own as a swap/temp drive if that would be faster for
say the OS swap file, \temp folder, and Photoshop's temp directory
too.

The swapfile isn't used anyway, so there is no gain from putting it on
another drive.
(If your swapfile is used a lot you should buy more memory instead of
RAID)

And what programs could be using a temp drive? The video editing
software is the reason why you want a raid array, so that stays on the
array.

But do you actually need all that disk performance?
Most reasonably intelligent editing programs don't need all that much
diskperformance

Marc

Photoshop uses a huge temp file (I sometimes find it has grown to 4GB)
to store all the cache history of image changes, and this slows things
down tremendously.

So I'm curious, everyone says RAID10 is faster than RAID5. I can fully
understand why for the same number of disks RAID0 is faster than
RAID5, but for the same number of disks (e.g. 4 in my case) RAID10
only has two disks in the stripe versus 4 disks in the RAID5 array.
Even with RAID5 being more overhead I'm surprised that having
effectively 4 disks versus 2 in the stripe of a 4 disk RAID10 array
doesn't match or exceed the performance. What kind of performance hit
does RAID5 cause versus RAID0 - it must be at least a 50% decline for
this to be the case.

Thanks,
Roland

 

[J. Clarke]

So, in XP, using a Promise Serial ATA S140 SX4 (with 256MB cache) and
four disks, 160GB 8GB, 7200rpm, either the Seagate or Samsung (looking
for quiet) which will I get the best performance from: either the 4
disks in a single RAID5 array, or the four disks in a RAID 10 array.
[Yes I know I'm talking 480GB versus 320GB of usable space in the end,
but performance is the question here]

Note that this is likely to be the only disk array in the system, so
would have the OS (WinXP) video editing, photoshop image editing etc
all on the same array. The disk system is most intensively used in
video and image editing.

That's easy. RAID10 is much faster.
Especially since you will also be writing to the disk.

RAID5 only performs well when sequentially reading large files, but is
slow when writing files.

Alternatively of course I could put 3 disks in the RAID 5 array, and
one disk on it's own as a swap/temp drive if that would be faster for
say the OS swap file, \temp folder, and Photoshop's temp directory
too.

The swapfile isn't used anyway, so there is no gain from putting it on
another drive.
(If your swapfile is used a lot you should buy more memory instead of
RAID)

And what programs could be using a temp drive? The video editing
software is the reason why you want a raid array, so that stays on the
array.

But do you actually need all that disk performance?
Most reasonably intelligent editing programs don't need all that much
diskperformance

Marc

Photoshop uses a huge temp file (I sometimes find it has grown to 4GB)
to store all the cache history of image changes, and this slows things
down tremendously.

So I'm curious, everyone says RAID10 is faster than RAID5. I can fully
understand why for the same number of disks RAID0 is faster than
RAID5, but for the same number of disks (e.g. 4 in my case) RAID10
only has two disks in the stripe versus 4 disks in the RAID5 array.
Even with RAID5 being more overhead I'm surprised that having
effectively 4 disks versus 2 in the stripe of a 4 disk RAID10 array
doesn't match or exceed the performance. What kind of performance hit
does RAID5 cause versus RAID0 - it must be at least a 50% decline for
this to be the case.

That would be true if RAID performance scaled linearly as a function of the
number of drives, but it doesn't. A four drive RAID doesn't have twice the
performance of a two drive RAID and a two drive RAID doesn't have twice the
performance of a single disk.

 

[Shailesh Humbad]

In a four-disk RAID-5, effectively only three disks are striped,
because one drive always gets the parity data. But three is still
better than two, so you may get better sequential read performance,
and especially random-read performance.

However, the logic breaks down for write operations. For each write,
the controller has to XOR the data values in all the stripes to
calculate the parity before writing to the disk. You need a dedicated
processor on the controller to handle that high volume of processing.
Worse, if a partial stripe is written, then the existing data in the
stripes has to be read and the parity recalculated before anything can
be written, which slows things down tremendously. I'm running WinXP
off a 4-disk RAID-5 right now, and the read performance is fantastic,
but the write performance is average, so I'm going to switch it over
to a RAID-10 or 2-disk RAID-0.

See these benchmarks:

http://www.somacon.com/fasttrak_sx4/page19.php

I'll add a benchmark against a RAID-0 sometime soon.

 

[Eric Gisin]

In a four-disk RAID-5, effectively only three disks are striped,
because one drive always gets the parity data. But three is still
better than two, so you may get better sequential read performance,
and especially random-read performance.

It should read from all four if the I/O request is large enough.

However, the logic breaks down for write operations. For each write,
the controller has to XOR the data values in all the stripes to
calculate the parity before writing to the disk. You need a dedicated
processor on the controller to handle that high volume of processing.

Nope, any system from the last 3 years can easily do it. Many people buy old
SCSI RAID cards only to find they are slower than Win 2K/XP RAID 5.

 

[Marc de Vries]

So, in XP, using a Promise Serial ATA S140 SX4 (with 256MB cache) and
four disks, 160GB 8GB, 7200rpm, either the Seagate or Samsung (looking
for quiet) which will I get the best performance from: either the 4
disks in a single RAID5 array, or the four disks in a RAID 10 array.
[Yes I know I'm talking 480GB versus 320GB of usable space in the end,
but performance is the question here]

Note that this is likely to be the only disk array in the system, so
would have the OS (WinXP) video editing, photoshop image editing etc
all on the same array. The disk system is most intensively used in
video and image editing.

That's easy. RAID10 is much faster.
Especially since you will also be writing to the disk.

RAID5 only performs well when sequentially reading large files, but is
slow when writing files.

Alternatively of course I could put 3 disks in the RAID 5 array, and
one disk on it's own as a swap/temp drive if that would be faster for
say the OS swap file, \temp folder, and Photoshop's temp directory
too.

The swapfile isn't used anyway, so there is no gain from putting it on
another drive.
(If your swapfile is used a lot you should buy more memory instead of
RAID)

And what programs could be using a temp drive? The video editing
software is the reason why you want a raid array, so that stays on the
array.

But do you actually need all that disk performance?
Most reasonably intelligent editing programs don't need all that much
diskperformance

Marc

Photoshop uses a huge temp file (I sometimes find it has grown to 4GB)
to store all the cache history of image changes, and this slows things
down tremendously.

I haven't had this issue. Perhaps a different configuration for
photoshop would help. Do you need 4GB of cached history of image
changes?

So I'm curious, everyone says RAID10 is faster than RAID5. I can fully
understand why for the same number of disks RAID0 is faster than
RAID5, but for the same number of disks (e.g. 4 in my case) RAID10
only has two disks in the stripe versus 4 disks in the RAID5 array.
Even with RAID5 being more overhead I'm surprised that having
effectively 4 disks versus 2 in the stripe of a 4 disk RAID10 array
doesn't match or exceed the performance. What kind of performance hit
does RAID5 cause versus RAID0 - it must be at least a 50% decline for
this to be the case.

The first thing to understand is that the performance increase from
RAID is only partly due to the increased transfer rate. Most
performance increase comes from the increased performance when reading
or writing random data.

When you look purely at the time it takes to write a 300MB file, then
the transfer rate will be the most important factor. But Raid5 is by
definition quite slow when writing data, because of the extra parity
calculations it needs to perform, compared to Raid1/0+1
Some people have already explained that. You can also see a good
explanation here:
http://www1.us.dell.com/content/topics/global.aspx/power/en/ps3q99_raid?c=us&l=en&s=esg

When reading a single large file, the Raid5 array is faster then
Raid10. (However many people will run into the limit of 32 bit 33 Mhz
PCI slots in their desktops, so that both will be equally fast)

In your case you will do both reading and writing. Probably even more
writing than reading. Raid5 is therefor at a disadvantage.

When reading smaller files, Raid arrays can have several disks each
reading a different small files. That is where the performance
increase for random read/writes comes from.
Transfer rate doesn't play a role in this situation. Raid 1 has this
capability without the need for striping. Since both disks have the
same information, you give each disk different files to read.
Because of this a Raid1 with 2 disks will have similar to a 2 disk
Raid0 array when reading smaller data.
Raid 10/0+1 is a combination of Raid0 and Raid1. So you can see that
it also has very good performance when reading data.

You can see a nice example of this in the following benchmarks:
http://www6.tomshardware.com/storage/20011023/raid-05.html

As you can see, the Raid0+1 arrays with 4 disks can perform about the
same as the Raid0 arrays with 4 disks. But it depends on the quality
of the controller. (The highpoints clearly are not that well
implemented) and the sort of activity.

Marc