Computer Hard Drive Configuration

[mwilson]

I am preparing to build a new computer and wanted some advice about the best hard drive configuration for photo editing. My computer case has 6 drive bays and I intend on using a PCIe based SSD as the boot drive, a second PCIe SSD as a scratch and Lightroom Catalog drive and 6 SATA drives for image file storage. I want to configure the SATA drives in RAID in order to have redundancy and improved write speed. After researching the various RAID options, it appears that RAID 10 or RAID 5 would be most appropriate for this application but although I understand the concepts used in both of these designs, I don't have a firm understanding as to how they would compare and contrast in a real world sense with respect to write speed and data protection. I presume that RAID 10 would be best configured as a mirror of a three drive stripe.
Intel i7980X, 16 GB 1600 DDR3, 6 ea. WD Black 1TB drives, 2 ea. OCZ R2 PCIe 512 GB drives

[MiniChris]

Just built mine a month ago...and it is a real hummer. Maybe this will get you started. I got all my stuff from www.newegg.com I would have included the prices, but everything pretty much changes day to day.

2G.SKILL Ripjaws Series 4GB (2 x 2GB) 240-Pin DDR3 SDRAM DDR3 1600 (PC3 12800) Desktop Memory
2Kingston DataTraveler G3 2GB USB 2.0 Flash Drive (White & Orange)
1Western Digital Caviar Black 1TB 3.5" SATA 6.0Gb/s Internal Hard Drive -Bare Drive
1Microsoft Wheel Mouse Optical
1Intel Core i7-870 2.93GHz LGA 1156 95W Quad-Core Processor
1SAPPHIRE Radeon HD 5450 100292DDR3L Video Card
1Microsoft Windows 7 Professional 64-bit 1-Pack for System Builders
1LITE-ON Black Wired Keyboard
1GIGABYTE gz-ph2a3 Black Computer Case
1Antec NEO ECO 620C 620W Continuous Power Power Supply
1GIGABYTE GA-P55A-UD3 SATA 6Gb/s USB 3.0 ATX Intel Motherboard
1ASUS 24X DVD Burner - Bulk Black SATA Model DRW-24B1ST/BLK/B/AS
I've since added a small mountain of additional RAM.

[mwilson]

That looks like a really great set up, but it appears that you have only one hard drive so, if that is the case, you aren't using RAID. If you have any experience with RAID, tell me about it. I have only used RAID 0 and RAID1 so I'm looking for advice from someone with experience with RAID 5 or 10.

[Colin Southern]

I use a couple of WD Raptors in a RAID 0 for my system drive on my processing PC, and a pair of 1.5TB in a RAID 1 for data. To be honest, I think you'll find that having a massive RAID setup like you're contemplating won't add much to the system's performance; RAM makes by far the biggest difference because RAID or no RAID, RAM is several orders of magnitude faster ... so sufficient RAM reduces paging to a minimum and buffers disk IO operations to the point where striped RAID arrays really won't give you that big an increase - infact, I'd be surprised if you'd see any significant improvement over a single drive for everything, unless you're doing something particularly IO intensive.

Also, last time I checked, a RAID 0 pair gave performance very comparible to a single SSD drive.

Hope this helps.

[Mario Xavier]

Let's break it down. Assuming all drives are 120GB

In a Raid 1 - you have redundancy. Everything that is written to one drive is written to the other drives in the pairing:
Drive A to Drive a - If either drive fails then the one remaining drive(s) will continue operating as if nothing happened. All data is safe until that drive fails.
Total storage in this setup is 120GB

In a Raid 0 - you have striping which is more of a performance boost. The drives split the load.
Drive A and Drive a will act as One. If you have two chunks of information Drive A will get one and Drive B will get the other but they will respond in unison. If either drive fails the other can't go on as it has the other half of the data. All data is lost.
Total storage in this setup is 240GB - (Both drives will act as a single 240GB drive)

In a Raid 10 (which is Raid 1 + 0) - You will have Three Raid 1s - Where data is striped across each Raid set.

Drive A and Drive a (1st Raid 1)
Drive B and Drive b (2nd Raid 1)
Drive C and Drive c (3rd Raid 1)

Drive A, B, and C split the load and then give an identical copy of their own load their clone.

So if we have 3 chunks of data
Drive A gets #1 and simultaneously copies that to Drive a
Drive B gets #2 and simultaneously copies that to Drive b
Drive C gets #3 and simultaneously copies that to Drive c
Total storage in this setup is 360GB

Combining knowledge of both Raid 1 and 0 to see how it stacks: The entire array will respond as ONE drive. As with Raid 1 - if Drive A fails then its clone (Drive a) will continue in it's place BUT since the overall array is striped (Raid 0) Drive a will still have only information on chunk #1 (as Drive A did). If it fails too, then Drives B and C are now without chunk #1 and can no longer make sense of the data they possess. All data is lost.

On the up side, Raid 10 can support a maximum of 3 simultaneous failures if 1 drive from each Raid fails. Drive A, B, and C could fail and Drive a,b,c, would continue striping. The worst thing that could happen is 2 simultaneous failures within the same Raid 1 configuration (Drive A and Drive a)

So if Raid 10 is what you want, make sure you have an extra drive that you can hot swap in the event of failure without necessarily powering down the system. That drive will immediately receive a copy of the data from the drive it's paired with and now your redundancy and integrity of the array is safe for the time being. I would hope that you would have a tool to help you locate the failed drive in a timely manner.

[Mario Xavier]

I agree with Colin here. I really don't think something of this magnitude is really necessary. When you get into Raid 10 you're looking at situations where there's going to be continuous and heavy amounts of reading and writing. We're talking small servers or something. For graphics and photos RAM will be the key player.

I have several drives in my computer. I use a program called Sync Toy from Microsoft to watch specific folders for changes and write them to an extra drive and to an external when my computer is idle. That gets the job done with money to spare.

[MiniChris]

I use Carbonite for my backup needs and a removable HD 1 terabyte...

[Ady]

I would agree with Colin and Mario that the suggested setup is a little over the top, given the performance of modern drives, however I would specifically recommend against using RAID 5 for this type of system.

Specific issues with RAID 5:
I would recommend avoiding a RAID 5 setup, principally for reasons of performance/cost ratio and support. A RAID 5 array will suffer from relatively poor write performance (particularly on small (less than a full stripe) or random writes) due to the read, modify, write cycle on both the data and parity. There is also an increase in latency on spinning disks as all the disks need to participate in the read, modify, write cycle. These issues can be largely mitigated by using a decent RAID controller with a hefty amount of battery backed write cache, and that can handle the parity calculations in the RAID hardware (many cheap and on-board RAID controllers are not true hardware RAID and do not support parity calculation in the RAID hardware and do not have a dedicated cache). Using SSDs will also help as there is no spin latency as the drive waits for the appropriate sector to be in position for read (or write) operations. The down side is that this type of setup will cost a small fortune without realising a significant real world increase in performance for the type of work you are doing. In addition to the performance issues there is also the problem that if your RAID controller gives up you may struggle to 'import' your RAID 5 array on to a new RAID controller so will be forced to rebuild the RAID and restore even if your drives are fine.

For internal storage I'd go with large capacity spinning disks, ideally enterprise class such as the Western Digital RE3 or RE4 drives for storage and a 10K RPM SATA (e.g. WD Velociraptor) for the system. If you really want to squeeze every ounce of performance out of your system then make sure you are x64 with a minimum of 8GB RAM to avoid system paging and add a pair of small SSDs as a RAID 0 for page/scratch/temporary storage. I would then use external drives (USB3 or eSATA) for backups.

I'm not generally averse to RAID and large collections of disks in RAID setups, my preference for storage is for storage enclosures with redundant everything (such as the Infortrend enclosures) on a fibre channel infrastructure, but that's a bit ridiculous for a domestic setup and a little excessive for many small businesses.

[mwilson]

I agree with Colin here. I really don't think something of this magnitude is really necessary. When you get into Raid 10 you're looking at situations where there's going to be continuous and heavy amounts of reading and writing. We're talking small servers or something. For graphics and photos RAM will be the key player.

I have several drives in my computer. I use a program called Sync Toy from Microsoft to watch specific folders for changes and write them to an extra drive and to an external when my computer is idle. That gets the job done with money to spare.

I think some background info would be appropriate here: On my current computer I have been using a few RAID 1 paired drives for image file storage and it has worked well for the past four years. But over time, the size of individual images had grown to the point that they often use 25 to 27 MB apiece up to 400 MB for some panoramas so read/write speed has become a problem. I normally make intermdiate file saves when working on an image and I want to reduce the time it takes to write to disc so I can speed up my workflow. I thought that using one RAID 10 instead of several RAID 1 configurations might improve write speed. I have noted that while waiting for the computer to save a file, my performance monitor typically indicates that the CPU is working at 10 - 20 % capacity and only 4-5 GB of my 8GB total is being used so I wanted to know about the relationship between RAM and write speed. You stated that RAM is the key player; should I overclock my RAM to improve perfromance?

[mwilson]

Thanks for the input, Colin. I would have a greater level of safety with three RAID 1 drives than one RAID 10 so RAID 10 doesn't make much sense if the read/write speeds aren't significantly faster than RAID1.

[Mario Xavier]

I wouldn't overclock it either. Overclocking anything depending on manufacturer suggestions usually lessens the life of it. The best thing I would do is make sure the bus speed of the motherboard matches the max speed of the RAM as well as note the cache size of the CPU. RAM can only operate as fast as the bus can transport data. In situations where the BUS is slower the RAM is underclocked. Overclocking it to run faster than whats allowed by the bus won't help the life of your components. However following the above recommendation, overclocking in this case leads to a true and often stable performance increase as all components are adequate to support a proportionate boost.

Another thing I'd like to point out is that RAID 10 is faster than Raid 1, how much is arguable. For the sake of a visual use the scenario of the 3 Chunks of data and exaggerate the write speed.

It takes 1 drive 3 seconds to right 3 chunks of data.

In a Raid 1 - it would take 3 seconds to write the data and clone it to the second drive.
In a Raid 10 - It would take 1 second to stripe the data and clone it. Because remember, the work is being split across the array.

The size of the cache of each drive significantly improves the read, write, and seek time individually thus overall "improvement" combined. Bleh...

I'd toss onto the table investing in a graphics card with plenty of RAM to take some of the load off of the CPU and system RAM. It will take care of the mathematical processing of your real time visual needs work, the CPU will take care of the interpreting the actions and the drives can focus on serving up the info, the memory can help each and relax. All components are complimenting each other and helping out. This is an efficient system.

Summary:
Plenty of cache (CPU, and Hard drives) (Increased virtual memory for the Hard drives also)
Hard drives rated for 24/7 use
Plenty of fast memory matching the BUS speed
A nice graphics card
Attached storage for backup

My system for example has to be able to support reading and writing of audio from 8 to 16 simultaneous audio inputs in wave format and display in real time when I'm working with bands. The effects applied are done in memory so for my situation I made sure that my system has the components necessary to help it handle each job. In working with audio my audio device handles all the input processing to take the load off of my CPU. Likewise the visual needs of graphic work should be split across a video card, memory, and cpu.

Make sense?

[mwilson]

I wouldn't overclock it either. Overclocking anything depending on manufacturer suggestions usually lessens the life of it. The best thing I would do is make sure the bus speed of the motherboard matches the max speed of the RAM as well as note the cache size of the CPU. RAM can only operate as fast as the bus can transport data. In situations where the BUS is slower the RAM is underclocked. Overclocking it to run faster than whats allowed by the bus won't help the life of your components. However following the above recommendation, overclocking in this case leads to a true and often stable performance increase as all components are adequate to support a proportionate boost.

Another thing I'd like to point out is that RAID 10 is faster than Raid 1, how much is arguable. For the sake of a visual use the scenario of the 3 Chunks of data and exaggerate the write speed.

It takes 1 drive 3 seconds to right 3 chunks of data.

In a Raid 1 - it would take 3 seconds to write the data and clone it to the second drive.
In a Raid 10 - It would take 1 second to stripe the data and clone it. Because remember, the work is being split across the array.

The size of the cache of each drive significantly improves the read, write, and seek time individually thus overall "improvement" combined. Bleh...

I'd toss onto the table investing in a graphics card with plenty of RAM to take some of the load off of the CPU and system RAM. It will take care of the mathematical processing of your real time visual needs work, the CPU will take care of the interpreting the actions and the drives can focus on serving up the info, the memory can help each and relax. All components are complimenting each other and helping out. This is an efficient system.

Summary:
Plenty of cache (CPU, and Hard drives) (Increased virtual memory for the Hard drives also)
Hard drives rated for 24/7 use
Plenty of fast memory matching the BUS speed
A nice graphics card
Attached storage for backup

My system for example has to be able to support reading and writing of audio from 8 to 16 simultaneous audio inputs in wave format and display in real time when I'm working with bands. The effects applied are done in memory so for my situation I made sure that my system has the components necessary to help it handle each job. In working with audio my audio device handles all the input processing to take the load off of my CPU. Likewise the visual needs of graphic work should be split across a video card, memory, and cpu.

Make sense?

That definitely makes sense. Thanks for taking the time to assist me. It has been a big help

[Colin Southern]

You stated that RAM is the key player; should I overclock my RAM to improve perfromance?

"Unfortunately" ...

... Microsoft have put a "safety valve" in the system - whereas is would be great if a system used all remaining available RAM for caching, and cached it for as long as was needed, in reality (off memory - it's been a while since I looked this up), I don't think they let anything in the cache age for more than about 5 seconds ... I guess the risk of power going off (or being turned off) is just too great (could result in the loss of lots of data). So the bottom line appears to be that at some point it does come back to the performance of your disk array.

Having said that though -- so long as your hardware is stable, and protected by a UPS, then I wouldn't get overly excited about saving too often. Personally I just save when I think about it -- usually when I've just nailed some kind of adjustment that I think I'd be struggling to get this close again. In reality, crashes are pretty rare. Also, 25MB files are pretty small; mine are often 10 times that size once adjustment layers are added and they don't take long to save (perhaps 10 seconds) (BTW, my system is Win7x64 - 12GB RAM - and an Extreme Edition CPU (with 8 logical cores). Keep in mind too though that the CPU utilization you're looking at is the whole CPU - it'll only ever be 100% when there are enough threads running to tie it up; although Photoshop supports multiple threads for some operations, I suspect that saving a file probably isn't multi-threaded, and thus is probably only using 1 core at any one time.

In theory, 3 or more drives in RAID 1 are safer than 2 drives in RAID 1, but that's a bit like saying a 747 is safer on long flights over water than a 767 because it has 4 engines -v- 2, but in reality, the chances of 2 drives failing at the same time is probably the similar to 2 engines failing at the same time ... in reality it's likely to be something else that takes out both of them (fire, software malfunction etc); no matter how many drives you have you STILL need to back them up to some reliable forms of external media, and get some of that media off site.

[mwilson]

I think we have a similar situation here. The file sizes to which I referred were as produced by the camera, in this case a Canon 1Ds Mk III, so the actual memory used during processing is considerably larger. So, having said all of this, how have you set up your storage drives?

Perhaps I'm paranoid, but I have been using a multi-tiered backup system that starts with RAID1 on chassis storage and is followed by a continuous backup to a RAID1 external drive (WD World Book II, 4TB) and also an assortment of external drives backed up every few days and taken off site.

[Sonic4Spuds]

Check these links,

http://www.tomshardware.com/forum/251893-32-raid-raid
http://www.tomshardware.com/forum/25...ence-raid-raid

Looks to me like the raid 5 is better for a higher storage capacity, while raid 10 is for hyper fast read write requirements.

-Sonic

[Colin Southern]

I think we have a similar situation here. The file sizes to which I referred were as produced by the camera, in this case a Canon 1Ds Mk III, so the actual memory used during processing is considerably larger. So, having said all of this, how have you set up your storage drives?

I'm using the same camera - and 16 bit workflows too of course. In terms of drives, all are connected directly to SATA ports on the mobo; the 2 Raptors are in a RAID 0 (for good system performance), and the 2x 1.5TB are RAID 1 for safety. I backup to an external 1TB HDD, but probably most significantly, commercial work is written to 2 DVD's - one I keep here at the workshop, and the other I keep in a warm / dark / dry place at an alternative location (big fancy words for "in my sock draw at home!).

Perhaps I'm paranoid, but I have been using a multi-tiered backup system that starts with RAID1 on chassis storage and is followed by a continuous backup to a RAID1 external drive (WD World Book II, 4TB) and also an assortment of external drives backed up every few days and taken off site.

You can never have too many backups

Just wondering if your continuous backup system is having any adverse affect on system performance?

[basfords]

I am currently using a Drobo. http://www.drobo.com/

Personally I didn't want to deal with Raid so this was a great option for me. Much easier to deal with. I can expand my storge with relative ease and if a drive dies my information is still safe. The website does a much better job of explaining how it works than I could dream of doing so I will let it speak for itself. I found out about it from a fellow photographer/computer programmer. I have been very happy with it so far. I hav enot been using it for very long so I can't really speak to the longevity of the system, but the person that told mne about it has been using it for years. He has had drives crash and the system worked as promised.

I believe there are a few different brands out there as well.

Sean

[steveedmonds]

I use something like drobo in raid 1 for backup, I regularly swap a drive I keep off site so if the house burns down, buy a new drobo, plug the drive in and your all back. The bakup is scheduled for I time I don't want performance from my machine.

If you really want disc write performance and reliability on your working machine have you thought of 15k SAS drives in harware raid 1.

steve

[McQ]

In terms of drives, all are connected directly to SATA ports on the mobo; the 2 Raptors are in a RAID 0 (for good system performance)

If you really want disc write performance and reliability on your working machine have you thought of 15k SAS drives in harware raid 1.

Hard drives? What are those? Might as well be recording your images on papyrus scrolls . I'd seriously consider moving to one of the latest SSD's for your boot drive (as opposed to a RAID 0). You'll get superior performance, in addition to better reliability as a bonus. Consider either the Crucial RealSSD 300, the recent Intel G2 models, or wait until the OCZ Vertex 3 is widely available in the coming weeks. They're dropping a lot in price, and it will likely be the single biggest boost you've given your computer in years...

The RAID 1 still makes tons of sense as a mass online storage area though, given the current price per GB of SSD storage.

[Colin Southern]

Hard drives? What are those? Might as well be recording your images on papyrus scrolls . I'd seriously consider moving to one of the latest SSD's for your boot drive (as opposed to a RAID 0). You'll get superior performance, in addition to better reliability as a bonus. Consider either the Crucial RealSSD 300, the recent Intel G2 models, or wait until the OCZ Vertex 3 is widely available in the coming weeks. They're dropping a lot in price, and it will likely be the single biggest boost you've given your computer in years...

Well maybe, but maybe not Either way, I'd love to see some comparisons. The latest Raptors have 32MB buffers - up to 6gb/s transfer rates from buffers - and a MTBF of 159 YEARS ... and that can then be put in a RAID 0!