RAID 0 performance?

[bobbydaz]

Just interested to know what performance boost I can expect by switching to a 2x 1TB RAID 0 set-up for my work files, System and Apps will be on a separate SSD.

Using mainly Photoshop (files 100mb up to max of 1gb) how much will my open and save times improve?

[Hellhammer]

RAID 0 theoretically doubles the read and write speeds. In real world, it depends on the task but it should speed the things up

[maflynn]

When I was running a RAID-0 setup, I noticed a significant decrease in boot times and program start up times, like photoshop.

If you're going with a RAID-0 setup, on your main drives, I'd be sure to have a solid backup regiment because if unit fails you lose all of your data.

[bobbydaz]

RAID 0 theoretically doubles the read and write speeds. In real world, it depends on the task but it should speed the things up

that's what I thought but haven't managed to find any test results to see what can actually be achieved.

[Hellhammer]

that's what I thought but haven't managed to find any test results to see what can actually be achieved.

Some benchmarks

http://www.anandtech.com/show/2212/8
http://www.anandtech.com/show/1371/5

I know they are using different drives but that's more to just compare RAID 0 and single HD. If you can't afford a big SSD, then RAIDing is you best solution. Just remember to have a solid backup as mentioned above

[bobbydaz]

Some benchmarks

http://www.anandtech.com/show/2212/8
http://www.anandtech.com/show/1371/5

I know they are using different drives but that's more to just compare RAID 0 and single HD. If you can't afford a big SSD, then RAIDing is you best solution. Just remember to have a solid backup as mentioned above

thanks for the links. I would like a large SSD but I though it wasn't a good idea to use SSD for constant writes, I would be writing data to it constantly 8 hours a day, 5 days a week!

[Loa]

Hello OP,

I just checked for Photoshop Open and Save times, using a single HD or a 4 disk RAID0. Test was made on a 2009 MP, with CS5, with 8GB of RAM. Test file is a 12MPx file with lots of complex layers/masks, and weighs 414MB.

The single HD is on an external eSATA dock, but you'll see that it makes no difference...

Open time:
-For single HD, External eSATA: 10 seconds
-For 4 disk RAID0: Internal: 10 seconds

Save time:
-For single HD, External eSATA: 40 seconds
-For 4 disk RAID0: Internal: 39 seconds

Bottle neck is clearly not the transfer speed! If you want faster Open/Save times, you need a faster CPU.

Loa

[Honumaui]

look into Raid 10 or 1+0 both the same ?

safe and speed with no controller to buy ?

400 meg file on my setup is about 3.7 seconds to open and save

[Loa]

look into Raid 10 or 1+0 both the same ?

safe and speed with no controller to buy ?

400 meg file on my setup is about 3.7 seconds to open and save

Nothing in your described set-up should be faster than a 4 disk RAID0 of enterprise grade drives.

What's the rest of your system? Saving a 400MB file in PS is always long, I'm curious how you managed 4 seconds

Loa

[Honumaui]

sorry about that :) here is my setup

actual file size was 396.5 megs


use a 100 gig SSD for boot
Areca 1222x for storage and I have older 750 gig HDDs in it ! not even faster new ones :) going to though next spring put some new ones in wait for the price of the RE4 to drop more
but its setup as Raid 6 also have the battery module on it
2 40 gig SSD in raid 0 for scratch
backup time machine is a standalone 5 disc raid 5 with 2 TB drives
the other is a JBOD PM case
then offsite storage as well

[bobbydaz]

Hello OP,

I just checked for Photoshop Open and Save times, using a single HD or a 4 disk RAID0. Test was made on a 2009 MP, with CS5, with 8GB of RAM. Test file is a 12MPx file with lots of complex layers/masks, and weighs 414MB.

The single HD is on an external eSATA dock, but you'll see that it makes no difference...

Open time:
-For single HD, External eSATA: 10 seconds
-For 4 disk RAID0: Internal: 10 seconds

Save time:
-For single HD, External eSATA: 40 seconds
-For 4 disk RAID0: Internal: 39 seconds

Bottle neck is clearly not the transfer speed! If you want faster Open/Save times, you need a faster CPU.

Loa

Thanks for the figures. A bit worrying that it sounds like it makes no difference at all. Am I wasting my time and money on a RAID 0 set up?

[Loa]

Honumaui: I can't understand how Photoshop can open or save a 400MB file in 4 seconds, even with your set-up. Is it a PSD file with lots of complex layers/masks?

I can save such a file to JPEG in a few seconds, but not as a complex PSD file. It's always long as there are many operations that PS does to it before even writing it to the disk.

Bobbydaz: I'm not sure about the real world applications of RAID0 unless you have to move a lot of really large files many times per day. There are probably a few apps that see significant gains (video work, most likely), but aside from that I've been underwhelmed by actual performance gain. Benchmarked throughput is impressive, but actual time saved in your typical workflow is entirely application dependent. Doesn't seem to help me with PS CS5.

Loa

[Honumaui]

PSD its about 7 layers ?

[bobbydaz]

PSD its about 7 layers ?

Just to throw something else into the mix - Would a single Velociraptor be a better option than RAID 0?

[Honumaui]

Just to throw something else into the mix - Would a single Velociraptor be a better option than RAID 0?

I might say yes if its one of the latest 600GB ones ? but without testing ? who knows unless someone else out their has ? and that depends on the raid 0 ?
if its two WD black 2TB shortstroked I might give the lead back to the 2TB raid 0 short stroked WD blacks then ?

the other thing in real world a short stroked WD 2TB would not be bad as the rest of the HDD can be used for BU ?

but if you have a raptor around ? you could try it ?

[Honumaui]

Hello OP,

I just checked for Photoshop Open and Save times, using a single HD or a 4 disk RAID0. Test was made on a 2009 MP, with CS5, with 8GB of RAM. Test file is a 12MPx file with lots of complex layers/masks, and weighs 414MB.

The single HD is on an external eSATA dock, but you'll see that it makes no difference...

Open time:
-For single HD, External eSATA: 10 seconds
-For 4 disk RAID0: Internal: 10 seconds

Save time:
-For single HD, External eSATA: 40 seconds
-For 4 disk RAID0: Internal: 39 seconds

Bottle neck is clearly not the transfer speed! If you want faster Open/Save times, you need a faster CPU.

Loa

after you open that file for grins go down to the lower left of the document and see what the scratch size is ?
and what was the efficiency ?

[Loa]

after you open that file for grins go down to the lower left of the document and see what the scratch size is ?
and what was the efficiency ?

Size: 1.12G/5.19G
Efficiency: 98%

Loa

[Honumaui]

Size: 1.12G/5.19G
Efficiency: 98%

Loa

thats not bad then for opening
you can also set your info window to show your efficiency and such might be interesting to keep a eye on it when you are working find where things get slower or use scratch etc.. ?
remember its just for the last thing done ? so complex actions wont record the total action just the last thing done :) but it can show a lot as you are doing stuff :)

[Loa]

Hello,

I know about scratch performance. What I still don't understand is how your system can save a 400MB PSD file in 4 seconds. I don't have the fastest set-up on the planet, but a 2009MP with a SSD boot drive and a 4 disk RAID0 set should be up there...

Getting 40 seconds while you get 4 seconds is a huge difference.

Loa

[nanofrog]

Hello,

I know about scratch performance. What I still don't understand is how your system can save a 400MB PSD file in 4 seconds. I don't have the fastest set-up on the planet, but a 2009MP with a SSD boot drive and a 4 disk RAID0 set should be up there...

Getting 40 seconds while you get 4 seconds is a huge difference.

Loa
The difference is a combination of the Areca's cache and he's running 8x disks in the array. ;) So even with older (slower) disks, his array ends up capable of a faster sustained throughput.

For example, if you figure 85MB/s per disk in the level 6, it should generate ~510MB/s, while 4x 100MB/s HDD's in a stripe set only generate 400MB/s. Rather obvious.

As per why it comes up 4x faster than your system, is yours is a software implementation (no cache or separate processor, though the actual RAID load should be very low). But it's still consuming system resources that could otherwise be used on the processing (assuming the test file is the same, which I'm thinking there may be more differences than seem evident from the information provided).

To give you an example, if I disable the cache in the ARC-1231ML (8x member RAID5), it can generate ~760MB/s sustained. But with the cache enabled, it can jump to ~1.4GB/s (file fits within the 2GB cache). So cache can make a significant difference.

[Honumaui]

Ditto Nanofrog on why

I also have 24 gigs of memory ? raid 0 2 disc SSD for cache and SSD for boot and Areca raid with 8 discs like NanoFrog mentioned

I dont check compatability ? that can add time on save ?

think of it this way the Areca or other true good raid cards have their own processor and their own memory and are why some of us preach once you want to get serious you look into this ? the downside is price ?
starting point $1000 without discs

a 400 meg file is pretty small when it comes to throwing them at the raid controller :)

and when I get the 1880 next season and see what discs are ripping then :) it will only get better :)

[Loa]

Hello,

I did think about that RAID card's cache. On the other hand, if throughput is the key to PS saving speeds, why does it take the exact same amount time to save the same file on an external eSATA single HD than it takes on my 4 disks RAID0 set?

Loa

[nanofrog]

Hello,

I did think about that RAID card's cache. On the other hand, if throughput is the key to PS saving speeds, why does it take the exact same amount time to save the same file on an external eSATA single HD than it takes on my 4 disks RAID0 set?

Loa
From what I can tell, you're a tad short on RAM, so the system is consuming cycles for paging (remember, you've the OS, PS, and RAID functions sharing resources on your system simultaneously).

The RAID card can take over the RAID functionality (i.e. more to do with getting the completed data off the RAM and into the cache on the card than the clock cycles used for the RAID calculations, which aren't much in terms of CPU utilization % for stripe sets). This frees up RAM to help workflow.

In your case, there's not quite enough from what I can tell (going by the 98% efficiency for PS), so it's paging RAM to/from disk (adds latency to completing the write you're measuring). Honumaui's system is running at 100% IIRC from a previous post made.

Can you borrow any RAM to get a larger capacity and retest?
If so, you can see if that 2% increase makes a real world difference in your system.

[Honumaui]

Hello,

I did think about that RAID card's cache. On the other hand, if throughput is the key to PS saving speeds, why does it take the exact same amount time to save the same file on an external eSATA single HD than it takes on my 4 disks RAID0 set?

Loa

could be if you have checked compatibility mode ? this will slow things down :) about %20 or more ?

it very well could be our file ? is yours in 16 bit mode ? that will do it ?
or the ram ? as mentioned ?

it is a bit of a jump ? for a 400 meg file ? I agree :)

for grins go here
http://forums.macrumors.com/showpost.php?p=10983737&postcount=76
do that ? down the thread #103 was my time and this was on my older 3,1 mac with only 14 gigs ? have not done it on my new one yet ?

would be cool to see your times and post em up :)


cause it could very well be your file will be close on my setup ?

I like figuring stuff like this out so game if you are to do some testing :)

save is throttled by one CPU !!!! sadly but if you can dump a bunch on that cache and then have it deal with it quick its going to help I found ? but again this much ? I think we are missing something between us maybe ?

for grins in testing that PS file my read is 2527.2 MB/s
checking disable cache it was 2763.4 MB/s

but the more I think ? their is something going on we are not catching :)

[Honumaui]

OK for grins I took the instructions and made a few changes
1. Open a new doc, 20inches x 20 inches i at 300ppi and color mode RGB 8 bit white background color profile sRGB

2. Filter -> Noise-> Add Noise -> Gaussian, 100%

3. Duplicate that layer 3 times to give you a total of four.

4. Make the top layer multiply, the next one down screen mode, the next one down, overlay.

5. Don't flatten. Save as a .PSD do not check compatibility mode when saving

I end up with a file that is 413.8 MB on disc

my open time was 4.1 seconds
save time was 3.8 seconds

I also have Lightroom and a ton of stuff open will do it from a fresh restart in a bit ?

this might help us get to the bottom :) and I really hope your time is 4 seconds :) cause it should be faster than what it is ? again I think we have a missing link :) so this would be fun to see what others get also for open and save :)

[Loa]

this might help us get to the bottom

Indeed:

Open time: 4 seconds
Save time: 4.8 seconds.

A bit slower than you, but now it's really close.

What's disappointing for me is that Open and Save times from my single HD eSATA drive are exactly the same. And since Photoshop is the app that works the hardest, I'm not sure I really benefit from RAID...

Loa

P.S. And yes, that original file is in 16bits. :-)

[Honumaui]

Indeed:

Open time: 4 seconds
Save time: 4.8 seconds.

A bit slower than you, but now it's really close.

What's disappointing for me is that Open and Save times from my single HD eSATA drive are exactly the same. And since Photoshop is the app that works the hardest, I'm not sure I really benefit from RAID...

Loa

P.S. And yes, that original file is in 16bits. :-)

yeah was curious about that :) 16 bit thing :)

glad its about the same :) was working and not thinking as much earlier ? then got the kids home from school sat down and read again and thought HMMMMMM that is a HUGE gap and had to think 16 bit or something ;) heheheh

bet you are feeling a lot better now :) and like me thinking MAN adobe has to rewrite PS from the ground up to take advantage of the corese better !!!!!



for grins take that test file and convert it to 16 bit
then save it as a TIF with no compression
the no compression shows up on the tif options near the top ;) after you save location

when done I get a 1.3 GB file it
3.8 seconds to save
4.0 seconds to open


save as PSD I get 573 MB file on disc
141 seconds to save
23.8 seconds to open

[Loa]

Hello,

Results for my 4 disk RAID0 followed by my single eSATA single HD:

Tiff file:
Open: 5 - 4.5 secs (no difference)
Save: 4.5 - 15 secs (huge difference)

PSD files:
Open: 15 - 16 secs (no difference)
Save: 85 - 91 secs (slight difference)

Very interesting results... It seems quite clear that for me (I only use the PSD and JPEG formats), a software RAID0 isn't any help with Photoshop.

Note: these results are the first time I open/save it. If I try it a few times without quitting PS, the times get longer and longer. (Is that because of my relatively low RAM?)

Loa

[Honumaui]

I would say yeah your ram is getting eaten up and or your scratch more your ram ? guess though :)

that last one that psd was 16 bit ?


I think the raid helps ? for me I have a PS job of 100+ images so if I can shave 3-5 seconds moving from one to the next it ads up I might do a few jobs a week like this ? other jobs run about 50 files and lots of single images or just a couple ? type jobs

if we do about 600 in layout those seconds ad up also in creating files the program creates a PSD then builds it and saves it


it also helps if I dont want to work on originals to quickly duplicate the directory and such

[Loa]

that last one that psd was 16 bit ?


Yep. I simply saved the 16bit TIFF file to PSD. It was still 16bit.

But in saving the PSD file I noticed something very interesting. My external dock has a disk activity light, and it only lit at the very end of that long saving time.

I don't know how fast your MP is, but maybe that could explain the difference in speed. I have a 2.66 2009 Quad.

Loa

[nanofrog]

Yep. I simply saved the 16bit TIFF file to PSD. It was still 16bit.

But in saving the PSD file I noticed something very interesting. My external dock has a disk activity light, and it only lit at the very end of that long saving time.

I don't know how fast your MP is, but maybe that could explain the difference in speed. I have a 2.66 2009 Quad.

Loa
I take it this was over an eSATA port then?

That could be a bottleneck, as it's been reported that SIL3132 cards for example aren't all that quick.

A complete detailed setup description could help find some of the issues. ;)

[jwestpro]

well guys, I did some of my own "photoshop saving" tests. nothing matter enough to change my day by much except for processor speed itself.

having a couple computers to play with and various source and destination options, i came up with the following. because actual time used is not always as useful, i have included the data transfer rate info from the "activity monitor"

1.84 Gb PSD file, 11 image layers + 11 adjustment layers, 16 bit. (on G5 in cs3, 2.7 Gb /2.38gb, 79% eff, 96 s open; on mp 6 core/3.33, 27s, 100% efficiency)

A. g5 dual 2.0 from 2003 or so, with 8gb ram, cs3:

1st save - via original SATA I on-board, direct to SATA I raptor 10k rpm drive-
85mb/s avg rate, 6:45, yes, that is almost 7 minutes ;)

2nd - via Sonnet internal sata II card to wd green 1T sata II 7200 hdd-
85-93mb/s, 6:20, 12 threads for what it's worth ;)

B. 6 core 3.33 machine with 24gb ram (3x8), CS5 with 80% allocated. software is running from 200 gb ssd via sata II board connection.
File is on separate ssd.

1st save to internal hdd sata II, 7200 rpm, 2:00, 112 mb/s peak transfer rate, this hdd is mostly empty and new.
2nd - Esata, through pcie card to raid 0 2T set 75% full, 1:55, 143 mb/s peak, 110-116 mb/s avg rate
3rd - internal single ssd where I have OS and apps, direct sata II cable in lower optical drive bay, 1:50, 138mb/s peak rate, 98-110 avg
4th - internal raid 0 set, 3x100gb ssd (owc RE extreme versions), loaded in first three main drive slots, connected to main board via sata II.
1:45 @ 138-157mb/s peak transfer rate.

So, while the below results show fast sequential write ability, it does us photoshop users no good at all. All we can hope for is faster Ghz processors and/or new code in the next cs6.

Using AJA read/write utility, 2gb file size, sustained writes (all ssd are owc "extreme pro RE" with the 28% provisioning):
481 mb/s to the 3x 100gb ssd, raid 0 internal, through mac board sata
395 to 2x 100 gb ssd
235 to 1x 200 gb ssd
138 to the esata raid 0, 2 disk lacie box via sata II and esata III card (newertech 6g)
89 mb/s to the internal sata II hdd, both 7200 rpm (1 is WD RE4, the other stock from apple)
37 mb/s to external FW 800 mini drive ;)

Ethernet transfer is a mystery, I thought it should be faster with direct connected cat 6 cables. moving a large file over from one ssd to another ssd in the laptop only hits 70mb/s.

[Honumaui]

Yep. I simply saved the 16bit TIFF file to PSD. It was still 16bit.

But in saving the PSD file I noticed something very interesting. My external dock has a disk activity light, and it only lit at the very end of that long saving time.

I don't know how fast your MP is, but maybe that could explain the difference in speed. I have a 2.66 2009 Quad.

Loa

this was on the 3.2 quad ? the other was on my 2.8 8 core ?

the light thing PS crunches then writes at the very end ? strange others have noticed this behavior as well

might retest my 16 bit one :)

[Honumaui]

Using AJA read/write utility, 2gb file size, sustained writes (all ssd are owc "extreme pro RE" with the 28% provisioning):
481 mb/s to the 3x 100gb ssd, raid 0 internal, through mac board sata
395 to 2x 100 gb ssd
235 to 1x 200 gb ssd


for grins here were some times I did the other day
with 3 of the 40 gig testing the 4 gig on ICH
475.7 write
553.8 read

the 3 40 gig on my areca were
665.7 write
696.8 read

2 using 4 gig on ICH
418 write
456.8 read


shows how throttled the ICH really is when you see what they can do on the Areca !

also yeah Photoshop really needs a complete make over !

[Loa]

I take it this was over an eSATA port then?

Yes, over eSATA. But I'm not surprised that it takes longer on a single external HD than on my internal RAID0! I'm glad it does!!! :-)

I was surprised that opening/saving that 16bit PSD file was faster on my software RAID0 than on Honumaui's set-up.

@JWestPro: if you're using TIFF files, then switching to a faster drive set-up (RAIDing) will probably help. Otherwise, only processor speeds seem to make significant differences.

Loa

[Honumaui]

Loa

for grins start it from 16 bit ?


1. Open a new doc, 20inches x 20 inches i at 300ppi and color mode RGB 16 bit white background color profile sRGB

2. Filter -> Noise-> Add Noise -> Gaussian, 100%

3. Duplicate that layer 3 times to give you a total of four.

4. Make the top layer multiply, the next one down screen mode, the next one down, overlay.

5. Don't flatten. Save as a .PSD do not check compatibility mode when saving


the psd how big was that on the disc mine is 633.4 megs
strange its a new size when I started from scratch than converting the tiff ?


even on my SSD Raid 0
23 seconds to open both times
144 to save first time
166 second time to save

[jwestpro]

@JWestPro: if you're using TIFF files, then switching to a faster drive set-up (RAIDing) will probably help. Otherwise, only processor speeds seem to make significant differences.

Loa

One of the tests above was already on a raid. And as far as I knew, I had to use PSD in order to retain all layers info.

However, I just read through all the Adobe help info on saving files and file types which says the PSD is really only necessary to maintain all the layers and data for OTHER applications. It also saves a composite so that other apps can "see" the file in a flattened form even if they cannot edit layers.

Furthermore, the help section says the TIFF is fine up to 4Gb and retains all layer information.

So, saving these large working files with all the layers in case I want to go back, has always been the bane of my photoshop existence. Now, a fully layered tiff version saves onto my 3xssd in 6 seconds.

I'm not sure what to say! Is there any archival reason I should be using the PSD format? I do, and would only ever, use photoshop to re-open these going back to rework something so I cannot see how the TIFF format is a problem at all.

[Honumaui]

yeah large tiffs with no compression preserve layers is a time saver :)

the only reason I use PSD is some clients who need the layered PSD docs ? and the only reason I check compatibility is when I don't know what flavor of PS they are on :)


and as noted since its pure drive speed not having to do anything heavy quicker HDD help ?

[Honumaui]

jwestpro would be curious to see your times with that test file thing I setup ? in both tif and psd for grins :)

[nanofrog]

Yes, over eSATA. But I'm not surprised that it takes longer on a single external HD than on my internal RAID0! I'm glad it does!!! :-)
For some reason, I thought the external was a 4x disk stripe set as well, not a single disk. :eek: So many different RAID threads, I occasionally get the various setups mixed up. :o

[Honumaui]

For some reason, I thought the external was a 4x disk stripe set as well, not a single disk. :eek: So many different RAID threads, I occasionally get the various setups mixed up. :o

you are doing better than me I do it every time I check the forum :)

[nanofrog]

you are doing better than me I do it every time I check the forum :)
:p It really seems to depend on the day (and how much re-reading I'm willing to do before posting = best defense from making an obvious mistake).

Unfortunately, my brain seems to want to run on Auto Pilot today for some reason. ;) :D :p

[bobbydaz]

Okay, just installed my new OCZ Vertex 2 60gb SD + 1tb x2 RAID 0 and decided to run a few quick tests.

MAC PRO 2.93 Quad/OSX 10.5.8/12gb Ram


PHOTOSHOP - open and save 320mb layered file:

Single HDD: 5.1 (open) 6.8 (save)

SSD: 5.2 (open) 5.2 (save)

RAID 0 HDD: 5.1 (open) 5.0 (save)


FINDER - copy a 320mb Photoshop file

Single HDD: 7.5

SSD: 3.2

RAID 0 HDD: 1.5

So to me it appears as if drive choice make no difference to open & save in Photoshop but for copying files RAID 0 beats SSD. Are these results as expected?

[Transporteur]

So to me it appears as if drive choice make no difference to open & save in Photoshop but for copying files RAID 0 beats SSD. Are these results as expected?

Seems a little strange to me. Copying files should be faster on the SSD.
Your RAID should be good for 200 to 220MB/s depending on the drives. I reached about 220MB/s with two 1TB WD Blacks. The SSD, however, should make at least 250MB/s sustained.
Where are your source and target directories when you copy?

SSD to SSD and RAID to RAID or are there always other drives involved?

[bobbydaz]

Seems a little strange to me. Copying files should be faster on the SSD.
Your RAID should be good for 200 to 220MB/s depending on the drives. I reached about 220MB/s with two 1TB WD Blacks. The SSD, however, should make at least 250MB/s sustained.
Where are your source and target directories when you copy?

SSD to SSD and RAID to RAID or are there always other drives involved?

Source and target were the same for each test, SSD to SSD etc. I agree I was surprised at the SSD performance. I will try copying some more files other than Photoshop and see if that makes a difference.

[bobbydaz]

Okay just ran the test again but with a couple of different file formats:

FINDER - copy a 900mb TIFF file

Single HDD: 10.8

SSD: 7.5

RAID 0 HDD: 9.4


FINDER - copy a 1.2gb EPS file

Single HDD: 28

SSD: 6.2

RAID 0 HDD: 6.3


Not sure what to make of these results, but it seems that with TIFFs, SSD is faster, but EPS about the same. Also why does the EPS take an incredible amount of time on single drive?

EDIT: have just run AJA tests..

Single HDD: 94-95 mb/s

SSD: 245-252 mb/s

RAID 0 HDD: 246-248 mb/s

[bobbydaz]

Can anyone confirm if my test results are normal?

[Honumaui]

hard to say :) but seems about inline with things ?

the one thing you could do is create a PS document like the one above adjust the size and give instructions then we could compare the exact file in a sense :)

not sure why the one on the HDD was so much longer ? how full was the HDD how fragmented did it go on and get written back to etc.. to many variables to ever know :)