Scratch Disk Choices and Sizing for new Mac

[Rod330]

First, I'll apologize in advance for the next newbie question about storage allocation. I've spent hours looking for the answer so hopefully I didn't miss an obvious post.

My new 2010 hexacore Mac will have 24G memory, an OWC Extreme Pro 120G SSD (positioned in the lower optical bay) and four 2TB WD RE4 drives. I'll use my old Synology 209 NAS (two 1TB WD Green Raid 1 drives) for TM backups over 1000Mbps Ethernet. Yes, I'll need to upgrade to a larger NAS very soon. This will be my first Mac Pro (have MBPs and iMac) and will be used for web design and development (mostly Adobe CS5 products).

The SSD will be used for boot and application files. I plan to use Disk Utility to create a 1+0 Raid array for data files. For a Scratch disk, should I:
1) create a partition on the 1+0 Raid Array for Scratch
2) partition a chunk of the SSD for Scratch
3) attach a spare external 2.5" WD drive via Firewire 800 for Scratch

Based on my reading, it seems that option 1 makes the most sense but I'm not entirely sure if you can partition a 1+0 array with Disk Utility. I'm pretty sure option 3 is quite silly but wanted to toss out the idea.

Lastly, are there any generic recommendations on scratch volume sizing? Thanks in advance for your time.

 

[Rod330]

After reading this post from nanofrog in the JulianBoolean thread, it seems that I'm on the right track:

Option 2:
Use an SSD as a OS/applications disk in the empty optical bay (boot/applications disk)
4x Mechanical disks in HDD bays 1 - 4, in a RAID 10 configuration (speed is ~ that of a 2 disk mechanical stripe set, but has the redundancy of 2x disks before data is gone). This is used for scratch and primary data (working files).

Any advice on scratch volume sizing guidelines? Thanks!

 

[Honumaui]

the scratch for PS will depend on how big your files are your memory (your case 24 gigs) and what you are doing to your files

trying not to hit scratch is key !
read this part
https://forums.macrumors.com/showthread.php?p=10995187&#post10995187

The SSD will be used for boot and application files. I plan to use Disk Utility to create a 1+0 Raid array for data files. For a Scratch disk, should I:
1) create a partition on the 1+0 Raid Array for Scratch
2) partition a chunk of the SSD for Scratch
3) attach a spare external 2.5" WD drive via Firewire 800 for Scratch

I would just point it at the 1+0 for a bit and see how the efficiency scratch thing is going ? for files I am rough guessing here under 1 gig you are going to be OK this is why our next layout machine arriving tomorrow I am putting 24 gigs in


if you end up being below that %95 mark a lot then I would say get a eSATA card and put a SSD on it or two HDD and partition off the top part and create a raid 0 for your scratch ?

to answer your other ones

1) why not partition off a piece of your raid ?
well lets use the analogy of a file cabinet if you had to put away papers fast would you want to keep having to open the top drawer all the time just to put away a small piece of paper once in a while ? or would it be better to do it all in one drawer !

remember you have to close the one drawer open the top put away close it open the lower and continue think of that analogy

2) I would not partition off your boot ? its better to keep it dedicated %100 you can do this but its not the best setup even though you are not doing much with your boot drive when working in PS ? you want at least 40 gigs or so and that might start crunching into breathing room for your OS ?
so YES you can but its not the best ?
but if you are hitting your scratch some ? might as well try it test it ? you might find for you and your setup it will work

3) never put a scratch on a slow HDD the fastest HDD that is not being accessed by another program is best

the one thing PS is really good at writing scratch as writing your data back in large writes ! and with big files this is one place a dedicated scratch can help



hope this helps

 

[bzollinger]

I'm not sure of the exact location of where on his site, but digilloyd I believe feels that it's best to put the scratch on the fastest drive. In your case and his, that means the SSD. Look at his site http://macperformanceguide.com/index.html under "setting up your mac" you should be able to find it there. Or look at the article about optimizing PS.

Right now I've got my scratch set to a RAID 0 2TB array and I don't notice any real speed increases.

Good luck!
BZ

 

[Honumaui]

I'm not sure of the exact location of where on his site, but digilloyd I believe feels that it's best to put the scratch on the fastest drive. In your case and his, that means the SSD. Look at his site http://macperformanceguide.com/index.html under "setting up your mac" you should be able to find it there. Or look at the article about optimizing PS.

Right now I've got my scratch set to a RAID 0 2TB array and I don't notice any real speed increases.

Good luck!
BZ

check if you are hitting your scratch much ?

check your efficiency if its every below %100 you are hitting it then look at the scratch number if the left is bigger than the right you are hitting scratch ?

depends on memory and size of images and what you are doing with them etc.. some hit them more than others ?

 

[Rod330]

Honumaui- thank you so much for the valuable feedback!

Bzollinger- thank you as well. Yes, I read the Digilloyd article but did not think using an SDD for scratch was a good idea because the frequent write activity would quickly deteriorate the SSD. I thought a fast mechanical disk (raid 0 or 1+0 perhaps) would be more appropriate even though it's not as speedy as the SSD. On the other hand, who am I to question Digilloyd's logic? Perhaps with the advent of the lower cost/lower capacity SSD's, this is much less of an issue if they essentially become "disposable" commodities after a couple of years.

 

[VirtualRain]

I would definitely use an SSD for scratch if you have enough capacity on your existing SSD or budget to buy another.

I would not partition an SSD... that's really unnecessary and may inhibit the drive controllers ability to maximize wear levelling across the drive.

As you say, for some applications, where high writes are likely to impact lifespan, you can almost consider them disposable assets, if the performance benefit vs cost makes sense for you. That's how I tend to look at my SSD's. I'd rather run them balls-out handling all my storage needs rather than baby them, and if I wear them out, I'll just replace them with something better, since there are new better drives coming out all the time.

 

[bzollinger]

check if you are hitting your scratch much ?

check your efficiency if its every below %100 you are hitting it then look at the scratch number if the left is bigger than the right you are hitting scratch ?

depends on memory and size of images and what you are doing with them etc.. some hit them more than others ?

I'd like to try this test, and have but can't seem to get all the details to do it properly.

Can you outline or point me to a step-by-step guide on how to check efficiency and/or if I'm hitting scratch?

thanks!

 

[stephenwales]

Stripe scratch disk across multiple disks

I've found that creating a striped volume and setting it as my CS5 scratch disk has worked wonders.

After following a tip regarding partitioning to take advantage of the faster outer 20-25% of each disk I've found that Photoshop even with only 4Gb ram is blazingly fast against compared to using a single disk or even the boot volume as most people leave as default.

I have 12Gb ram so don't access the scratch unless im working on a large Taxi or Vanwrap but I would recommend this to anyone

Ive attached a quick image of my setup and the breakdown of partitions
Volume1 is the fastest as its using the outer rim of each platter which is the fastest with Volume2 holding current work documents and Volume3 holding personal stuff like music and films with archived Work Documents

Macintosh HD, Volume2 & Volume3 backup to 2x 4Tb (2drive) FW Raid in Lacie Boxes on alternate days using Carbon Copy Cloner and also 24/7 to Backblaze (mostly because I'm 101% paranoid)

 

[Honumaui]

having the same discs have partitions that are accessed the same time as data is not the best ? yes it can work but also work against you ?

if you are writing large files this will be slower it was in my tests in the past and I was using up to 8 discs in raid 0 and everytime it was the same or slower again large files write you are hurting yourself

I would put a bet on if you did some actual time tests that just pointing the scratch at the same setup as your documents is going to be as quick as what you have ?

best analogy you now have a two drawer file cabinet that your head has to go up to the other partition then back to the data traveling longer distances being slower PS is very good about using disc space near your data as scratch

not saying what you have is not faster than your previous ?
but the way you have it optimized is not the best IMHO ?

a single SSD for scratch is going to beat what you have easy every time all the time !

having a outer partition where your data starts but includes that scratch so say the outer %35-%40 of the discs and using that for data and scratch I bet if you time will end up being the same or quicker sometimes and will use the space better ?
writing huge files your heads are now going to have to be trying to do scratch and write files which will be way slower

the only time it will/might be quicker maybe is in a few instances with certain filters or things as nothing else is going on no writes etc.. but then the time saved might be lost back ?

again I would test as everyone has dif workflows etc. if it works cool but again I might think from my own experience you are kinda going back and forth and loosing consistency

if you work with larger files I would say change for sure

if working with smaller files ? you might be fine keeping the setup you have as the write loss will be minor and since the speed will balance out most likely you are going to be OK with what you have

but again test heavily the complete open close workflow as its always the best idea to never have discs trying to access two partitions at the same time in any workflow


just some thoughts

I've found that creating a striped volume and setting it as my CS5 scratch disk has worked wonders.

After following a tip regarding partitioning to take advantage of the faster outer 20-25% of each disk I've found that Photoshop even with only 4Gb ram is blazingly fast against compared to using a single disk or even the boot volume as most people leave as default.

I have 12Gb ram so don't access the scratch unless im working on a large Taxi or Vanwrap but I would recommend this to anyone

Ive attached a quick image of my setup and the breakdown of partitions
Volume1 is the fastest as its using the outer rim of each platter which is the fastest with Volume2 holding current work documents and Volume3 holding personal stuff like music and films with archived Work Documents

Macintosh HD, Volume2 & Volume3 backup to 2x 4Tb (2drive) FW Raid in Lacie Boxes on alternate days using Carbon Copy Cloner and also 24/7 to Backblaze (mostly because I'm 101% paranoid)

 

[stephenwales]

Thanks for that I'll try a change of Scratch location to test

I was under the impression from Adobe however that once photoshop had read a file from disk it did not refer back to that file until you saved it, that all work took place in ram and on the scratch disk. This is again why I made the scratch disk the outer fastest rim of my disks.

Wouldn't be the first time Adobes own advice was wrong! Any thoughts?

having the same discs have partitions that are accessed the same time as data is not the best ? yes it can work but also work against you ?

if you are writing large files this will be slower it was in my tests in the past and I was using up to 8 discs in raid 0 and everytime it was the same or slower again large files write you are hurting yourself

I would put a bet on if you did some actual time tests that just pointing the scratch at the same setup as your documents is going to be as quick as what you have ?

best analogy you now have a two drawer file cabinet that your head has to go up to the other partition then back to the data traveling longer distances being slower PS is very good about using disc space near your data as scratch

not saying what you have is not faster than your previous ?
but the way you have it optimized is not the best IMHO ?

a single SSD for scratch is going to beat what you have easy every time all the time !

having a outer partition where your data starts but includes that scratch so say the outer %35-%40 of the discs and using that for data and scratch I bet if you time will end up being the same or quicker sometimes and will use the space better ?
writing huge files your heads are now going to have to be trying to do scratch and write files which will be way slower

the only time it will/might be quicker maybe is in a few instances with certain filters or things as nothing else is going on no writes etc.. but then the time saved might be lost back ?

again I would test as everyone has dif workflows etc. if it works cool but again I might think from my own experience you are kinda going back and forth and loosing consistency

if you work with larger files I would say change for sure

if working with smaller files ? you might be fine keeping the setup you have as the write loss will be minor and since the speed will balance out most likely you are going to be OK with what you have

but again test heavily the complete open close workflow as its always the best idea to never have discs trying to access two partitions at the same time in any workflow


just some thoughts

 

[Honumaui]

Thanks for that I'll try a change of Scratch location to test

I was under the impression from Adobe however that once photoshop had read a file from disk it did not refer back to that file until you saved it, that all work took place in ram and on the scratch disk. This is again why I made the scratch disk the outer fastest rim of my disks.

Wouldn't be the first time Adobes own advice was wrong! Any thoughts?

why I say test hehehehe and tried to say it in a way where I am not saying I AM CORRECT !@ I hate that kinda attitude and try to say it with more test yourself this is what I have found with my tests and why

but yes you are correct and adobe is correct but the problem is you have to open and close files and that is where the major slow down is ?
unless you work on one or two files all day and really dont care about opening and closing you might be best with what you have

I think peoples perception of workflow speed is what they feel ? meaning
I HATE waiting any time for anything but opening and closing is not a big deal as my brain sets me up for OK I am opening or closing BU say applying a filter where times might be waiting .25 or .5 for something else ? cause its that break in the workflow that makes me scream so to say ? does that make sense what I am trying to say
so for me it might be the open and close loose that extra 1 second is worth it to have something take .25 rather than .5 cause that .25 second makes me scream

but then again I say get both as fast as you can best of both worlds

so I guess if a file opens in 3 seconds or 5 seconds you save that 2 seconds with dedicated ? again depends on file size and the same for writes ?

so the setup you have will it be quicker if the savings using yours vs one partition (again being on a 4 disc raid 0) if your times are that much faster with that scratch a bit farther out on the platter ? so instead of a filter taking 2 seconds it now takes 2.1
do yo do enough of those to make up that extra write or close time

so again I say test for you needs and file sizes

and really mentioned this just as another viewpoint and to make people think HMMMM

I love testing for myself cause you might find what you have is the best
I was not trying to say it was not
but I had tested that setup before and for me it was slower this was on regular platter HDD

but again now that I use SSD!!!! WOW for scratch and cache SSD are killer I know it must have mostly to do with seek and access times but might be worth looking into if you want that extra boost and your memory is not enough ? and in PS sometimes memory is short and nothing you can do about that ! so scratch then becomes a must have !!!

I now use a 2 disc SSD raid 0 setup for my scratch for PS

 

[nanofrog]

After reading this post from nanofrog in the JulianBoolean thread, it seems that I'm on the right track:

Please keep in mind, there's a vast number of ways you can go about RAID, depending on need and budget.

The quote you pulled up, was based on a very limited budget, so it's rather compromised.

If at all possible, separate the scratch data from your project data (separate drives or arrays), so they can be accessed simultaneously.

As Honumaui's mentioned, check out your memory usage, as that's extremely important (you want to keep scratch usage to as little as possible).

I would definitely use an SSD for scratch if you have enough capacity on your existing SSD or budget to buy another.

If at all possible, Yes, even though it will wear. OWC has smaller capacity drives that will fit the need, and are reasonably priced. Worst case, plan out a short MTBR (wouldn't go over 3 years, even if it were SAS disks), and toss them.

The actual wear imposed on the drive, is down to the OP's usage, and is impossible to predict ATM (no idea how much data is written to the scratch space per day). But by using cheap but still speedy SSD's, new options have opened up to users, as they can be considered as disposable as mechanical disks.

I would not partition an SSD... that's really unnecessary and may inhibit the drive controllers ability to maximize wear levelling across the drive.

I wouldn't recommend partitioning an SSD either, as it will almost certainly have a negative impact on wear leveling.

As you say, for some applications, where high writes are likely to impact lifespan, you can almost consider them disposable assets, if the performance benefit vs cost makes sense for you. That's how I tend to look at my SSD's. I'd rather run them balls-out handling all my storage needs rather than baby them, and if I wear them out, I'll just replace them with something better, since there are new better drives coming out all the time.

Exactly. Some of the more recent models from OWC can be used like this, as they're not terribly expensive (i.e. smaller capacity versions, as not much will be needed if the memory is configured sufficiently = reduced writes to scratch anyway).

 

[JulianBoolean]

I'd like to try this test, and have but can't seem to get all the details to do it properly.

Can you outline or point me to a step-by-step guide on how to check efficiency and/or if I'm hitting scratch?

thanks!

No problemo. You need the look at two things, preferably at the same time.

1. in photoshop show the info pallete. use the pull down menu on the info pallete (top right side) to get to the prefs. make sure you have efficiency and scratch size checked.
you'll also see the price paid per number of history states you've chosen to save. every move you make gets saved to a history state. that stuff gets held in ram, until the ram runs out. some things use oddles of ram, liquify filter, rotate, scale, convert profile etc.

2. open your activity monitor, and click on the system memory tab. the green area shows available ram. you'll notice the green area goes away the more moves you make in photoshop, or open more programs, etc.

My thread about the DX4 also has four screen grabs showing how to monitor ram usage. see page two. that thread btw, has slowly gotten gotten pretty deep into configuration options for a 14k - 19k workstation. It's kind of part II of the retouching thread linked here previously.
https://forums.macrumors.com/threads/1030834/

Hope this helps!

- Julian

 

[Honumaui]

one thing also for PS
if you have complex actions sometimes its good to put a history state then a purge command in their to help free back up memory and clear out the history states etc.. long complex actions I do this for