Pages: 1 2 [3]   Go Down

Author Topic: Scratch Disk size and speed  (Read 33458 times)

Christopher

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1499
    • http://www.hauser-photoart.com
Scratch Disk size and speed
« Reply #40 on: March 13, 2010, 06:36:55 pm »

Quote from: feppe
I wasn't referring to SSDs in storage sense, but OS and scratch disk sense. But I guess some people can afford to use SSDs for storage...

A few perhaps ;-), I would prefer to upgrade my whole computer every year for the same money ;-)
Logged
Christopher Hauser
[email=chris@hauser-p

Schewe

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6229
    • http:www.schewephoto.com
Scratch Disk size and speed
« Reply #41 on: March 14, 2010, 01:15:10 am »

Quote from: Jack Flesher
PS for Jeff: I would reconsider the SAS drive array.  SATA3 is already here, and in a very short while you'll be able to load your box up with 4 @ 2TB SATA3 drives for probably less than the 4 smaller SSDs are going to cost, and it will likely be a lot faster. Just sayin...


Yeah, well the whole point of my system upgrade was to do it once and forget about it for 3 years (which is what I did last time). I know some SATA 3's are out there but I don't think they will be as fast as 1%K SAS drives...so I'll deal with it. All my images and LR cataloges will actually be on an external 6 drive stripped array (backed up to a duplicate 6 drive stripped array) that will be giving me sustained 350-400 MB/second...

and considering the 6 drives will be 2TB, that means I'll have just under 12TBs of backed up on-line storage that should be pretty fast...

So, of course, I order my MacPro then the rumors start about the dual 6-core boxes...ya know what? don't care...I just know my dual quad-core from 2007 is pretty darn slow and I need an upgrade...
Logged

Jack Flesher

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2592
    • www.getdpi.com
Scratch Disk size and speed
« Reply #42 on: March 16, 2010, 12:06:45 pm »

Well, different strokes... .  

Personally, I think 4 of these in R-0 would be pretty freaking kick-butt even in my old 8-core box: http://www.newegg.com/Product/Product.aspx...N82E16820148349

Logged
Jack
[url=http://forum.getdpi.com/forum/

Christopher

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1499
    • http://www.hauser-photoart.com
Scratch Disk size and speed
« Reply #43 on: March 16, 2010, 07:54:30 pm »

Quote from: Jack Flesher
Well, different strokes... .  

Personally, I think 4 of these in R-0 would be pretty freaking kick-butt even in my old 8-core box: http://www.newegg.com/Product/Product.aspx...N82E16820148349

As long as you completly delet them every month it should be nice, otherwise as scratch these SSDs in RAId would be a quick desaster. ( RAID = NO TRIM )
Logged
Christopher Hauser
[email=chris@hauser-p

Jack Flesher

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2592
    • www.getdpi.com
Scratch Disk size and speed
« Reply #44 on: March 17, 2010, 09:09:17 am »

Quote from: Christopher
As long as you completly delet them every month it should be nice, otherwise as scratch these SSDs in RAId would be a quick desaster. ( RAID = NO TRIM )

Not for scratch.  For folks that want an uber-fast working image I/O storage tank.  As I indicated earlier in this thread, I do NOT recommend SSD's for scratch because of the wear issue, and scratch is exactly the type of operation that clogs SSD's.  As a working image storage pot though, they would be the performance kings, easily outperforming an SAS array at not too much difference in cost once you factor in the SAS controller.  (Actually, IIRC AnandTech did a comparison about a year ago and a pair of SATA2 SSD's in R-0 are about the same speed as a 4x R-0 SAS array in a sustained read or write, and smoked the SAS on random I/O -- SATA3 SSD will be better still...)  

The downside for SSD's remains though --- depending on how much I/O you actually do, you'd want to scrub (recondition) an SSD array on some regular schedule, probably at least quarterly, to keep performance up...  And one simple wipe and reformat won't do it, you need to repeat that about 4 times to reset the drive's write controller chip.  So if you factor in the time required for this maintenance across your overall usage, things get muddier.  Personally, if I was after performance,  I'd still go with the SSDs over an SAS array, but that's me...

Cheers,
« Last Edit: March 17, 2010, 09:27:24 am by Jack Flesher »
Logged
Jack
[url=http://forum.getdpi.com/forum/

Schewe

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6229
    • http:www.schewephoto.com
Scratch Disk size and speed
« Reply #45 on: April 18, 2010, 02:18:33 am »

Quote from: Jack Flesher
Well, different strokes... .


Thought I would follow up and let people know that I do have my new Mac Pro installed and "broken in"...

I have my main storage external in a pair of Burly Systems 6 bay in a RAID 0 (stripped) array...I have two of these and RAID 1 is backed up to RAID 2 nitely...I also have a Drobo Pro that get an occasional backup from RAID 2.

For the internal Mac Pro drives I have 4 600 gig SAS drives. I have my boot on a single drive and my Users folder on a second drive. I also have a RAID 0 stripped array as my Photoshop CS5 scratch drive. These SAS drives are running off of an Apple RAID card...the dual quad core, 2.93 GHz main CPU plus 32 gigs of ram are very, very fast for Photoshop CS5 (I'm running Ps CS5 with 70% allocated which works out to 21818MB for Photoshop).

The Burley Arrays are giving me about 350MB/sec through put and the SAS Boot and User drive is at up to 426 MB/sec for reads and 142 MB/sec writes...the scratch array is up to 434 MB/sec reads and 289 MB/sec writes (which really helps Photoshop for the rare times I hit scratch).

So far my only issue is the heat...I've had to add a fan under the deck to help get rid of the heat that is generated...it isn't so bad when the weather is cold but Chicago has recently had days in the 80's and the heat generated is really noticable (I have to work with my shoes off).

Rumor has it Apple will be coming out with new towers in July and SATA 3 and SSDs might be worth looking into in the near future but as of now, I'm happy with the new system and Photoshop CS5.
Logged

Jack Flesher

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2592
    • www.getdpi.com
Scratch Disk size and speed
« Reply #46 on: April 18, 2010, 09:34:00 am »

It sounds like you've built an awesome hot-rod Jeff, congrats on the new machine!
Logged
Jack
[url=http://forum.getdpi.com/forum/

Adam L

  • Full Member
  • ***
  • Offline Offline
  • Posts: 220
    • http://adamlozo.com
Scratch Disk size and speed
« Reply #47 on: April 21, 2010, 11:54:25 am »

Off subject question for J. Schewe:   I noticed that you've selected a relatively unknown (to me) DAS by Burly Arrays.   I visited their web site as I'm in process of upgrading my system.   I like the dual PM system http://www.burlystorage.com/ccp0-catshow/burlydualpm.html  but am experiencing sticker shock as this is more than the competitors.   What drove you to this solution?  I'd purchase a single system and raid 0 three drives and use the other three as a backup.

On another side note, I had an interesting call with the Burly support guy who is also a 20 year practicing photographer with Lightroom experience.   He suggested that I stay away from the 600GB SATA 6.0Gbps 10000RPM - 3.5" - Western Digital VelociRaptor in a Raid 0 pair.  He said that it's a 2.5 inch disk and that I'd get better scratch performance using a single 1TB SATA 7200RPM - 3.5" - Seagate Barracuda® 7200.12.    The reason is that storage works from the outside of the disk and that a larger sized disk with more space would out perform the smaller platters.   I don't know if this is correct or not but I did find this statement to have a true ring to it.

I keep tweaking my system specs as I learn more about what works well with Lightroom and Photoshop.  By the time I decide I suspect that 6-core systems will be the norm.  Oy

Logged
"That's a lot of money to move a few pix

Schewe

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6229
    • http:www.schewephoto.com
Scratch Disk size and speed
« Reply #48 on: April 21, 2010, 01:03:07 pm »

Quote from: Adam L
Off subject question for J. Schewe:   I noticed that you've selected a relatively unknown (to me) DAS by Burly Arrays.   I visited their web site as I'm in process of upgrading my system.   I like the dual PM system http://www.burlystorage.com/ccp0-catshow/burlydualpm.html  but am experiencing sticker shock as this is more than the competitors.   What drove you to this solution?  I'd purchase a single system and raid 0 three drives and use the other three as a backup.

On another side note, I had an interesting call with the Burly support guy who is also a 20 year practicing photographer with Lightroom experience.   He suggested that I stay away from the 600GB SATA 6.0Gbps 10000RPM - 3.5" - Western Digital VelociRaptor in a Raid 0 pair.  He said that it's a 2.5 inch disk and that I'd get better scratch performance using a single 1TB SATA 7200RPM - 3.5" - Seagate Barracuda® 7200.12.    The reason is that storage works from the outside of the disk and that a larger sized disk with more space would out perform the smaller platters.   I don't know if this is correct or not but I did find this statement to have a true ring to it.

I keep tweaking my system specs as I learn more about what works well with Lightroom and Photoshop.  By the time I decide I suspect that 6-core systems will be the norm.  Oy


I've got the 600GB SAS 15k drives-they are indeed 3.5" not 2.5" and they cost a lot more than the 10K. And yes, my System and User folders are on two separate drives on the outside of the platters (first partition) as suggested by Rick. The internal 2 drive array is only for Photoshop scratch. All 4 drives are running off of the Apple raid card since it supports SAS drives and can be made to boot from either one of the drives connected to it or an arrayed pair connected to it.

The jury is still somewhat out regarding the use of two drives and the first partitions for faster small block random read and write access...It's not at all easy to compare and test this. Things seem real fast now compared to my aging old MacPro. But the only slightly daunting aspect is the fact it's "complicated" in that you have two separate places where stuff has to be and it's up to you to keep track and back it up.

As to the Burly Dual PM, at about $700 before drives, I don't know where else you'll find a 6 drive box...I've seen 4, 5 and 8 drive boxes but this is the only 6 bay I'm aware of. I compared this box to Cal Digit, G-Tech, Buffalo Station and a few others I think. This was the best overall deal when it came to the box, the drives, the card and the final price and capacity.

I've had a Burly raid 5, 4 bay box running 24/7 for 5 or 6 years on a server...once you get the stuff set up, it's durable. They ain't sexy looking...that's why I hide mine under the desk where all you can see is the blinking lights!

:~)
Logged

Janne Aavasalo

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 60
Scratch Disk size and speed
« Reply #49 on: April 22, 2010, 01:48:39 pm »

Hey all,

I just read trough the thread and one thing kept bugging me the whole way. This is a thing that I think only Jeff Schewe has taken into account (that I noticed).

There was a lot of talk about 7200 and 10000rpm drives and then the better option, which is SSD - drives, debate about Trim and different Raid - levels.

But no-one except Jeff mentioned a raid controller. I don't know Macs too well, but since they are Intel based these days, I'm pretty sure they are basically the same when it comes to on board sata controllers (which suck big time).

If you really want performance out of your raid arrays, the only way to fly is an "third party" controller card. Preferably from a good manufacturer (Areca comes to mind) and which support Raid 6.

Why Raid 6 when my mobo does Raid 5?

Raid 5 is quite prone to hard-drive failures. Lets say that you buy 4 disks for your array. You'll probably go with same size disks (right), then you'll want same kind of drives for your array (right or wrong, depending), but the last thing will be the thing that'll probably bite you in the backside of things in the future. You buy them from the same place at the same time (definitely wrong).

This is fine when everything is fine. But these days when hard drives are cheap, you'll probably go with 1-2Tb drives. The problem starts to build up.

The next part is when they start failing (not if). They are bought at the same time, so they are probably from the same batch. They have been used about the same amount of time, so when one of them fails and you replace it, the controller starts to build the array again. During this time the activity is quite high on the disks and with large arrays the rebuild time is so long, that you are lucky if another drive doesn't go out during the rebuild. This is where Raid 5 fails, in the middle of the rebuild, another drive fails and the whole array is gone.

Raid 6 patches this a bit, since two drives can go, so it's reducing the possibility of a total failure, but then again, Raid 6 support is quite minimal (you'll have to go with the controller card), you'll have to get more disks for this to work and finally, Raid 6 is slower than Raid 5.

On that note, fake raid systems (mobo/software raid) are not really good, since Raid 0 and Raid 1 are the only ones that work even close to what they are supposed to. Raid 0 does make the I/O faster, but not by too much. Raid 1 is slower than one disk by itself and Raid 5 performance is a bad joke.

I use fake raid (Raid 1 arrays only) on my backup server, because it doesn't have to be that fast. Only thing I want is some redundancy. I do have two SSD - drives on my main rig in Raid 0, but that's because I wanted them to be a bit faster and also I wanted the combined space of 2 x 64Gb for my operating system.

What comes to Trim support and the lack of it, my disks are quite early and cheap versions, which are without Trim and I cant "feel" any slowing down yet (been using this installation of Win 7 for about 5 months).

And if you don't store anything important to the disks (Photoshop scratch, operating system (which is non critical to me) etc.) you can do a re-format with a linux live disk, so that your disks are returned to the state when they were brand new. That should fix the lack of Trim if you do it often enough.

These were just my 0.02$, so feel free to comment.

Janne
Logged

Christopher

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1499
    • http://www.hauser-photoart.com
Scratch Disk size and speed
« Reply #50 on: April 22, 2010, 04:06:23 pm »

Well, RAID controllers are probably the most important thing. In my new workstation there are two, one very powerful one from LSI for my 5 x 2TB drives which make a RAID 5. Why RAID 5 ? Because it is just faster than RAID 6 and it is secure enough. I have another two external RAID 6 backup devices.

The second RAID controller is a Adaptec and uses RAID 0 with 4 x 500GB drives, which make up my scratch disk.

There are two more SSDs hanging on the mainboard, which are not used in any RAID, but as System and Lightroom drive.

I can say that this system is just amazing.  I made my decision against SAS drives because I don't see any huge benefit. I personally think that a combination of SSDs and normal SATA drives is the better choice.
Logged
Christopher Hauser
[email=chris@hauser-p

Janne Aavasalo

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 60
Scratch Disk size and speed
« Reply #51 on: April 22, 2010, 05:20:07 pm »

Hey Cristopher,

Sounds like a sweet system and it seems that you've made solid choices when building it.

I can understand your Raid 5 choice here well, since Raid 6 is somewhat slower system. But as a only means of backup or disk space without backup, I'd stay far away from 5.

Janne
Logged

Christopher

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1499
    • http://www.hauser-photoart.com
Scratch Disk size and speed
« Reply #52 on: April 22, 2010, 06:45:16 pm »

Quote from: JanneAavasalo
Hey Cristopher,

Sounds like a sweet system and it seems that you've made solid choices when building it.

I can understand your Raid 5 choice here well, since Raid 6 is somewhat slower system. But as a only means of backup or disk space without backup, I'd stay far away from 5.

Janne

Yes you are absolutely correct that as a backup it should be a RAID system which can manage two failures. As a Working RAID drive I prefer the speed of RAID 5.
Logged
Christopher Hauser
[email=chris@hauser-p

sty

  • Newbie
  • *
  • Offline Offline
  • Posts: 10
Scratch Disk size and speed
« Reply #53 on: May 18, 2010, 01:37:45 pm »

I've noticed these last 6 months while managing a 6-disk zfs (3x1TB, 3x1.5TB) system that raid5/6 is not reliable enough for storing data. Bit flips happen and raid5/6 (or whichever level) doesn't have a mechanism to check the data when it's read.

I scrub my disks on zfs once per week and almost every time there's at least a few bits changed here and there causing a 32 byte block to be reconstructed. I believe that currently zfs is the only production capable filesystem that has end-to-end data integrity (btrfs is years away from production quality).

So for actually storing data on big and cheap disks, I'd stay away from any raid system if you want to keep your files bit perfect.

I've actually noticed that my lightroom library doesn't have any performance problems residing on the server that would actually impact my workflow, performance is almost as good as on ssd (gigabit network of course).
Logged

PierreVandevenne

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 512
    • http://www.datarescue.com/life
Scratch Disk size and speed
« Reply #54 on: May 19, 2010, 04:29:40 am »

Quote from: JanneAavasalo
What comes to Trim support and the lack of it, my disks are quite early and cheap versions, which are without Trim and I cant "feel" any slowing down yet (been using this installation of Win 7 for about 5 months).

On my Intel SSDs, trim is definitely useful. I have several hundreds of thousands of e-mails in my e-mail database, basically everything I sent and received since 1993: my e-mail program takes a few seconds to start on a fast hard drive, and around 1 second on the SSD. After a month or so, the SSD is down to the speed of the hard drive. Running the Intel utility restores the near instant response of the SSD. Of course, this is an ideal scenario for the SSD as all those mails are organized in hundreds of folders which have their own databases and indexes. In order to display the folder structure and the mail headers, there are lots of small reads all over the place, and lots of small changes that can fragment stuff as I use the mail program.

Logged

Jack Flesher

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2592
    • www.getdpi.com
Scratch Disk size and speed
« Reply #55 on: May 19, 2010, 09:57:44 am »

Quote from: PierreVandevenne
On my Intel SSDs, trim is definitely useful.

Even nicer than trim IMO is the sandforce solution... Currently available on enterprise-level OCZ and OWC SSD's...
Logged
Jack
[url=http://forum.getdpi.com/forum/

Rob C

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 24074
Scratch Disk size and speed
« Reply #56 on: May 19, 2010, 10:26:35 am »

To think that none of this interchange would have been possible (or certainly would have been thought lsd induced) when I bought my first 500C. Happy, innocent days!

Rob C
Pages: 1 2 [3]   Go Up