Benefits of partitioning a SSD?

[sparkie7]

Is there any benefits to partitioning up say a 250GB SSD into say 4 partitions to:

1. improve read/write times since they are smaller "volumes" to seek and access? i'm thinking probably not on the fastest SSDs as they seem to be <.1ms

2. set-up a RAID 0. Is it possibly to strip a RAID using partitions on the same physical SSD? obviously there are advantages to stripping 2 or 4 physical SSDs, but wondered if there is any advantage to do it with a single SSD via partitions

3. separate files into: Partition 1: Sys & Apps; P2: Docs; P3 & P4: scratch disk

The reason I'm brining this up was that I was going to strip 2 x 120GB SSDs. But they are all sold out an only have one 250GB SSD. And thinking is there a way to optimise/better setup and utilise a larger drive

One last question, is it possible via Disk Utility to set up 3. above with p1 & p2. And then setup P3 & P4 as a RAID 0?

Finally should I push the button now and buy the 250 or wait for 2 x 120GB when new stocks arrive in a couple of weeks

Appreciate the advice guys

 

[Spanky Deluxe]

Raid on a single drive's usually a bit of a waste of time but might be different with SSDs. To be honest though, is this your first SSD machine? If yes then I'd suggest you get the 250GB SSD and use it as a single partitioned drive and then later on down the line when the drives become cheaper, get another 250GB SSD and do your RAID setup then.

 

[sparkie7]

Raid on a single drive's usually a bit of a waste of time but might be different with SSDs. To be honest though, is this your first SSD machine? If yes then I'd suggest you get the 250GB SSD and use it as a single partitioned drive and then later on down the line when the drives become cheaper, get another 250GB SSD and do your RAID setup then.

you mean chuck everything into the one drive: sys, applications and documents in one volume?

my other machine (MBP 17") has a 128 SSD, and I just use it as one volume

 

[Spanky Deluxe]

you mean chuck everything into the one drive: sys, applications and documents in one volume?

my other machine (MBP 17") has a 128 SSD, and I just use it as one volume

Yeah that's what I'd do, one volume. Then when you can afford it, get yourself another 256GB SSD, back you primary up and then raid the two together so as to have 512GB of lovely Raid 0 SSD storage. Why limit your upgrade path by going for two smaller drives now?

 

[sparkie7]

Yeah that's what I'd do, one volume. Then when you can afford it, get yourself another 256GB SSD, back you primary up and then raid the two together so as to have 512GB of lovely Raid 0 SSD storage. Why limit your upgrade path by going for two smaller drives now?

good point. but i was wondering how much scratch disk (as a RAID 0) I'd realistically need with photoshop.

Plus the 250GB SSD is 20M B/s slower than the 120 SSD in writes performance. In the overall scheme of things its negligible I guess. Plus I think the anticipated firmware update is supposed to level write performance across so they are the same write speeds

 

[Spanky Deluxe]

good point. but i was wondering how much scratch disk (as a RAID 0) I'd realistically need with photoshop.

Plus the 250GB SSD is 20M B/s slower than the 120 SSD in writes performance. In the overall scheme of things its negligible I guess. Plus I think the anticipated firmware update is supposed to level write performance across so they are the same write speeds

From what I've heard, the reliability of SSD drives means that you may as well make your whole two disk setup a RAID 0 array. Only you know how much of a scratch disk you really need. You could get a rough estimate by working out the uncompressed size of the files you work with (number of pixels * bit colour depth / 8) * the number of files you normally have open * the number of layers you usually use. If you were using 10 megapixel files with a 32 bit colour depth and usually had ten files open with up to ten layers each then you could estimate that you'd need about (10 * 1,000,000 * 32 * 10 * 10 / 8) 4,000,000,000 bytes or ~4GB.

If I were you, I'd use one big disk or if you have two disks, one big raid 0 array. Then create a "scratch disk" folder, set Photoshop to use that space for its scratch files, get a cheap 3.5" SATA regular hard drive and set Time Machine to back up your SSD drive, setting Time Machine so that it does *not* backup the scratch folder.

 

[sidewinder]

From what I've heard, the reliability of SSD drives means that you may as well make your whole two disk setup a RAID 0 array.

SSD's have an MTBF that is no better than enterprise class SATA drives. Don't get caught up thinking SSD's are somehow bullet proof.

S-

 

[Spanky Deluxe]

No, I would never assume they're bullet proof - but they're certainly meant to be far more reliable than normal drives largely because normal drives can get clogged up with stuff like dust over time which can cause the insides to mess up and break prematurely whereas SSD's are largely immune to such effects. I'd consider an SSD Raid 0 boot/application/data volume acceptably safe if it was regularly backed up to a regular hard disk drive via Time Machine.

 

[VirtualRain]

Is there any benefits to partitioning up say a 250GB SSD into say 4 partitions to:

1. improve read/write times since they are smaller "volumes" to seek and access? i'm thinking probably not on the fastest SSDs as they seem to be <.1ms

2. set-up a RAID 0. Is it possibly to strip a RAID using partitions on the same physical SSD? obviously there are advantages to stripping 2 or 4 physical SSDs, but wondered if there is any advantage to do it with a single SSD via partitions

3. separate files into: Partition 1: Sys & Apps; P2: Docs; P3 & P4: scratch disk

The reason I'm brining this up was that I was going to strip 2 x 120GB SSDs. But they are all sold out an only have one 250GB SSD. And thinking is there a way to optimise/better setup and utilise a larger drive

One last question, is it possible via Disk Utility to set up 3. above with p1 & p2. And then setup P3 & P4 as a RAID 0?

Finally should I push the button now and buy the 250 or wait for 2 x 120GB when new stocks arrive in a couple of weeks

Appreciate the advice guys

All of your strategies mentioned above are designed to overcome the limitations of seek times and limited I/O on magnetic drives.

SSD's effectively have near zero seek times which changes the game. It's no longer necessary to put different files on different volumes to optimize I/O. The SSD's are also fundamentally large RAID0 arrays of memory chips... that are already accessed in a highly parallel manner. Stripping on the same physical disk will result in zero gain. Finally, partitioning to create smaller volumes worked on magnetic drives to reduce the physical movement of the drive heads which is also no longer relevant with SSD's... an SSD can be 10GB or 1TB and access to any part of the volume takes exactly the same time (nearly zero).

Stripping across multiple physical SSD's still has plenty of merit and you will benefit from a RAID0 array of 2x120GB drives vs. a single 240GB drive. However, the advantages are likely only to be seen with very large file reads.

Another way to improve RAID (or even non-RAID) performance is to use a RAID controller with write back cache capability that can effectively free up the OS to go about it's business without waiting for the write to commit to disk. This will make small writes appear near instantaneous. OCZ has implemented write back cache in it's recent line of Vertex SSD's which has had a dramatic effect on write performance.

SSD's change the game and you need to think of it more like RAM than old magnetic disks with spinning platters and heads that move back and forth.

 

[sidewinder]

No, I would never assume they're bullet proof - but they're certainly meant to be far more reliable than normal drives largely because normal drives can get clogged up with stuff like dust over time which can cause the insides to mess up and break prematurely whereas SSD's are largely immune to such effects. I'd consider an SSD Raid 0 boot/application/data volume acceptably safe if it was regularly backed up to a regular hard disk drive via Time Machine.

Time out.

Hard drives are hermetically sealed. If they get dust in them you have much bigger problems. If you are going to talk about why hard drives fail, at least use real reasons.

SSD's are relatively new technology and being solid state does not mean they won't fail. If their MTBF was much higher than enterprise class hard drives they would have a much higher MTBF rating. They don't so they are not....

S-

 

[sparkie7]

All of your strategies mentioned above are designed to overcome the limitations of seek times and limited I/O on magnetic drives.

SSD's effectively have near zero seek times which changes the game. It's no longer necessary to put different files on different volumes to optimize I/O. The SSD's are also fundamentally large RAID0 arrays of memory chips... that are already accessed in a highly parallel manner. Stripping on the same physical disk will result in zero gain. Finally, partitioning to create smaller volumes worked on magnetic drives to reduce the physical movement of the drive heads which is also no longer relevant with SSD's... an SSD can be 10GB or 1TB and access to any part of the volume takes exactly the same time (nearly zero).

Stripping across multiple physical SSD's still has plenty of merit and you will benefit from a RAID0 array of 2x120GB drives vs. a single 240GB drive. However, the advantages are likely only to be seen with very large file reads.

Another way to improve RAID (or even non-RAID) performance is to use a RAID controller with write back cache capability that can effectively free up the OS to go about it's business without waiting for the write to commit to disk. This will make small writes appear near instantaneous. OCZ has implemented write back cache in it's recent line of Vertex SSD's which has had a dramatic effect on write performance.

SSD's change the game and you need to think of it more like RAM than old magnetic disks with spinning platters and heads that move back and forth.

Excellent summation. Thanks for the feedback

 

[Igotmacqs]

Finally should I push the button now and buy the 250 or wait for 2 x 120GB when new stocks arrive in a couple of weeks

Which SSD drives are you looking at buying? Where?

Thanks.

 

[sparkie7]

Which SSD drives are you looking at buying? Where?

Thanks.

I ordered a OCZ Vertex 250 GB SSD. the 120's were gone

 

[Spanky Deluxe]

Time out.

Hard drives are hermetically sealed. If they get dust in them you have much bigger problems. If you are going to talk about why hard drives fail, at least use real reasons.

SSD's are relatively new technology and being solid state does not mean they won't fail. If their MTBF was much higher than enterprise class hard drives they would have a much higher MTBF rating. They don't so they are not....

S-

I'm afraid they're not hermetically sealed. That's just a myth. Just look into some advanced hard drive data recovery info and you'll quickly realise this. MTBF times apply to hard drives that are kept in a clean environment and undero no physical external forces - i.e. never moved anywhere. The MTBF times for an SSD in a similar environment is comparable, however, under real world use, SSD drives are largely unaffected by these other things that can cause drives to fail before their MTBF times, i.e. external physical forces and dust.

 

[sidewinder]

I'm afraid they're not hermetically sealed. That's just a myth.

Okay, air is allowed to pass into and out of most hard drives. But, that air passes through a carbon filter that prevents any particles large enough to be an issue from passing through. The air that passes through does so because of changes in temperature inside the drive. There is no active air flow. Desiccant backs are in place to absorb any moisture.

So, for all intents and purposes, the drive are sealed against dust and any other particles that could cause failures. A dusty environment does not increase the chance of drive failure due to particle-induced head crashes.

S-

 

[schalliol]

I'm bringing this thread up again now that you can reasonably add ExpressCard SSDs to your MacBook Pros (well unless you just bought a 15"). While slightly off-topic, I just picked up a 48GB ExpressCard SSD that fits completely inside the ExpressCard slot on my unibody MBP (reviews) for $175 shipped from NewEgg to go with my 320@7.2k HDD. This SSD can read at 115MBps and write at 65MBps. I wondered if anyone had suggestions for the best configuration. I understand it would be a great documents disc and many people find they could be decent boot volumes. According to SoftRAID, you might even RAID it up for certain applications: "How our RAID driver works is queued reads, so it would do the best possible performance with mismatched performing devices. So I think SoftRAID would be fine in these cases, not being worse than the faster of each individual device. Also, if one were significantly faster than the other, you can set it to only read from one device. Writes are always limited by the slower of all devices, though." It would be interesting to swap the 320 with a SSD and RAID it up too. Thoughts? Thanks!

 

[Tesselator]

Is there any benefits to partitioning up say a 250GB SSD into say 4 partitions to:

1. improve read/write times since they are smaller "volumes" to seek and access? i'm thinking probably not on the fastest SSDs as they seem to be <.1ms

2. set-up a RAID 0. Is it possibly to strip a RAID using partitions on the same physical SSD? obviously there are advantages to stripping 2 or 4 physical SSDs, but wondered if there is any advantage to do it with a single SSD via partitions

3. separate files into: Partition 1: Sys & Apps; P2: Docs; P3 & P4: scratch disk

The reason I'm brining this up was that I was going to strip 2 x 120GB SSDs. But they are all sold out an only have one 250GB SSD. And thinking is there a way to optimise/better setup and utilise a larger drive

One last question, is it possible via Disk Utility to set up 3. above with p1 & p2. And then setup P3 & P4 as a RAID 0?

Finally should I push the button now and buy the 250 or wait for 2 x 120GB when new stocks arrive in a couple of weeks

Appreciate the advice guys

I think partitioning or RAIDing (or both) a single SSD won't affect speed much either way unless there's more overhead in Apple's software controller than I think. There's no moving parts in an SSD. Memory (file) locations are accessed at the speed of sending a few commands to the controller by the devices refresh rate. So to get to any physical location is going to be about the same speed. So I guess the only benefit would be the organizational aspect of doing so.

A RAID0 of two or more SSDs would be faster because the limited controller speeds (throughput) are multiplied in an interleaved fashion allowing the interface to be more richly saturated with data. Kinda like how two hoses can fill up a bucket faster than one hose can.

So a stripe set of 2 x 120GB would potentially be faster than one 250GB SSD.

@schalliol
Because of the symmetry or RAID0's interleaving I would assume transfer speeds for both read and write would be a combination of the devices respective speeds. Not as fast as the fastest or as slow as the slowest in either direction but rather faster at both R/W operations than either one device singly. Unless of course one was slower than the overall advantages of RAID0's data interleaving.

 

[300D]

A dusty environment does not increase the chance of drive failure due to particle-induced head crashes.

However, it will plug the vent and increase the chance of dust ingestion or overheating.

 

[Claus84]

Actually I think it might be a really bad idea, partitioning the drive?

I read this article at Anandtech yesterday about the Intel X25M SSD (http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3403&p=3) and how they used some kind of wear and tear algoritm to determine which cells have been written to most often and thereby avoiding using the same cells over and over again. They spread out the wear of the disk evenly - thereby extending the overall lifetime of the SSD.

I have no idea how partitioning works on a SSD but if it is anything like on a magnetic disk, partitioning would actually hinder this algoritm in working its magic... thereby shortening the lifetime of your SSD.

Anyone able to confirm this, as it's just a thought I had... ?

 

[300D]

Thats exactly what would happen. Basically, the life of the cells would be cut in half.

Notice the quoted expected lifespan of an SSD is longer as the drive size increases.

 

[Tesselator]

If that's true it would also seem to imply that an SSD filled to 75% capacity would also have a reduced life-span even without partitioning. Since I guess almost everyone with an 80Gig is operating at around 75% full then brace yourselves for impending doom.

 

[frimple]

Thats exactly what would happen. Basically, the life of the cells would be cut in half.

Notice the quoted expected lifespan of an SSD is longer as the drive size increases.

Oh I suppose I read that the exact opposite. Let's say you split a 80 gig Intel SSD in half (two partitions), used one and not the other. I would wager that the unused half is the prime candidate for wear leveling and is utilized just as the extra space is (Intel provides extra "space" on their drives for this purpose and cached writing). I submit that the firmware is partition intensive and doesn't care if this block is partition A and that block is partition B. All it sees is writes/block and which one is free.

 

[nanofrog]

If that's true it would also seem to imply that an SSD filled to 75% capacity would also have a reduced life-span even without partitioning. Since I guess almost everyone with an 80Gig is operating at around 75% full then brace yourselves for impending doom.

Yes, sort of. It's more of the remaining available cell count, than percentage. That is, a 75% full 32GB drive is in more danger (8GB unused) than a 75% full 256GB (64GB unused). The latter still has a greater quantity of unused cells for wear leveling. 8x in fact.

Partitioning amplifies this effect.

 

[Tesselator]

Yes, sort of. It's more of the remaining available cell count, than percentage. That is, a 75% full 32GB drive is in more danger (8GB unused) than a 75% full 256GB (64GB unused). The latter still has a greater quantity of unused cells for wear leveling. 8x in fact.

Partitioning amplifies this effect.

Yup, that makes total sense. So with SSDs the more you store on them the less reliable they become. The plot thickens!