Solild State Disks

[John Kotches]

This is a good question, and one that has been keeping me from investing in a Solid state drive for quite some time. To the best of my knowledge even the best Flash Memory has has a maximum write life cycle of about 100,000 writes per block. Good wear leveling can help extend the life of a device by readdressing regularly written to blocks, but if you have a program or OS that regularly pages data, this could severely limit the life of one of these drives. I was really hoping that MRAM (Magnetoresistive Random Access Memory - Wikipedia, the free encyclopedia) Would take off as it seems more resistant to write related wear, but it is still too expensive to be competetive with Flash.

Keep in mind that they are rating these devices into the Millions of Hours for MTTF -- and that the wear leveling is getting ever more complex.

If you have a program or OS that is regularly paging data, you have either a defective program (or OS) or a system with insufficient physical memory for the intended purpose of the system. Take your pick. In this era of multiple GBs of RAM using swap unless you've dramatically overloaded the machine is unnecessary. If you are finding large amounts of swap is occurring, utilizing a mechanical spindle or two strictly for swap and enjoy the high latency each time you page or swap in and out.

In a server based environment I'm concerned with the intent is to never have to "crack" the chassis during the service lifetime of the machine (3 years). In the next 6-12 months (probably as part of our DC migrations) we'll be moving all of our servers over to SSDs.

For big storage (think very large EMC / NetApp or comparable) we're still going to be using traditional storage for all but tier 0.

 

[John Kotches]

This is a good question, and one that has been keeping me from investing in a Solid state drive for quite some time. To the best of my knowledge even the best Flash Memory has has a maximum write life cycle of about 100,000 writes per block. Good wear leveling can help extend the life of a device by readdressing regularly written to blocks, but if you have a program or OS that regularly pages data, this could severely limit the life of one of these drives. I was really hoping that MRAM (Magnetoresistive Random Access Memory - Wikipedia, the free encyclopedia) Would take off as it seems more resistant to write related wear, but it is still too expensive to be competetive with Flash.

Another point to make, that the better SSDs (MLC no-less) have their wear leveling algorithms so highly tuned that they can write somewhere around 60% of the drive's capacity daily and not exceed the write thresholds on any single cell for 5-10 years.

At that point, the machine would be obsolete.

 

[Smith P.]

"If you have a program or OS that is regularly paging data, you have either a defective program (or OS) or a system with insufficient physical memory for the intended purpose of the system. "

In Windows world it wouldn't be that simple.
If you will read Mark Russinivich article about Windows swap file size, you'll find more aspects of the caching/paging/swapping/etc processes.

 

[John Kotches]

"If you have a program or OS that is regularly paging data, you have either a defective program (or OS) or a system with insufficient physical memory for the intended purpose of the system. "

In Windows world it wouldn't be that simple.
If you will read Mark Russinivich article about Windows swap file size, you'll find more aspects of the caching/paging/swapping/etc processes.

I consider Windows to be a defective OS for many reasons; not the least of which is the use of swap when there is available physical memory.

 

[navychop]

My first recollection of a discussion about the upcoming demise of the "spinning metal" goes back some 15 years, after a Jerry Pournelle article in BYTE.
We paid about $1/MB then and pay 1/10 of that for 1000 times more storage today. And the drives' performance is not even in the same league...

Around the time DVD was introduced, there was an almost consensus that the dual layer version will kill hard drives dead.

I wouldn't rush to write off hard drives just yet...

Diogen.

You remember that article, too, huh? Too bad I bought in to it. Sounded good at the time.

Remember bubble memory? Great idea, slow in performance. Another idea that was supposed to replace spinning disks.

 

[John Kotches]

Diogen:

I'm not writing them off yet.

I was however incredibly surprised at how quickly they are getting to competive sizes with main stream HDs. Just last year at CES we were excited about the potential of 64 & 128GB drives.

Now it's 512GB drives and one 1TB announced.

For "everyday" storage, I see SSD eclipsing mechanical spindles in a few years especially as prices continue to crater on the SSDs.

 

[mike123abc]

If you are building a new i7 platform, just put in 12GB+ of RAM and do not have a paging file. Then you do not have to worry about Window's lazy paging system.

 

[Smith P.]

Some SW will eat all RAM regardless the physical size - really, that Mark Russinovich article will help you understand better Win OS paging.

 

[navychop]

And RAM limitations do not reside solely in operating systems. Ever have a motherboard that capped your RAM? Limited # of slots, limited to single sided, limited address range, etc.

 

[Foxbat]

I have 10 GB in my Mac Pro. No problems here...

I would like to get a couple of the 64GB MLC SSDs for our SQL servers to house some of the indexes and tables that get the majority of hits. I'm leery, thought, of a device that's rated in millions of writes instead of 10^15. Are there any hybrid SSDs that use dynamic RAM for storage with a cheaper flash to back it up when the power is removed?

We had some DEC drives like that back in the 1990s, but they were only 160 MB eek and cost $10K or more. They used a conventional 3.5" IDE drive to back the RAM plus a NiCd battery to allow the RAM to be written to the HDD when the power was removed. I would think the modern version would allow Read/Write speeds that would approach the SATA/SAS limits.

 

[snathanb]

Title of this thread should be changed. SSD = Solid State Drive, not Disk. No disks involved.

 

[Stargazer]

If SSD are more susceptable to failure after a shorter period of time for continuous use then it would be nice to know far enough in advance with some software so that you know to replace it and save all the content in advance before you lose everything.

 

[mike123abc]

If SSD are more susceptable to failure after a shorter period of time for continuous use then it would be nice to know far enough in advance with some software so that you know to replace it and save all the content in advance before you lose everything.

The failure of SSDs is that writes no longer work. In theory your drive becomes read only. You do not lose any data. Hence the allure of SSDs.

 

[John Kotches]

If SSD are more susceptable to failure after a shorter period of time for continuous use then it would be nice to know far enough in advance with some software so that you know to replace it and save all the content in advance before you lose everything.

What are you defining as "continuous use" though?

As an example, Intel rates its drives at 60% of capacity written daily takes 5 years to go through the specified number of writes.

Also, I would be remiss if I didn't point out that it's not going to just stop writing after 10,000 cycles. It will last a minimum of 10,000 cycles reliably.

 

[diogen]

Unless a "killer application" is found (pronto!) to trigger massive increase in demand of 1TB+ drives,
it will remain a niche for the "server guys", i.e. the demand will remain low just the same as for raw CPU power lately.

Let's have a look at this pricing table (taken from here).

Let's assume regular hard drive technology advances stop.
SSD drives today are 50 times more expensive than "spinning metal" (see table: 7.44 vs. 0.14).
SSD is silicon, hence, price halves every 18 months, i.e. in 8 years it will catch regular HDDs.
Considering HDD will in fact improve, it would be safe to estimate 15 years as the time SDD will catch HDD in terms of price per GB.

Diogen.

EDIT: Most probably SDD will carve out a niche for itself, e.g. system drives.
But "Big Storage" won't be taken over by SDD for a long time to come, IMHO.

 

[John Kotches]

Diogen:

That Intel you cite (it's the X25-M 80GB) has dropped 50% in price in < 6 months. It was introduced at US$800 and is currently selling for $400.

The same is true for the other manufacturers.

I'm a firm believer in putting all of the big storage outside of the machine on protected storage. It's taken way too long to rip my content that's out there to lose it to a single drive failure.

 

[mike123abc]

I just installed on a 64 GB SSD. It was $134.99.

They are closing in fast on the $2/gb price. I am still not ready to install my main computers with them, but for special computers like this one, it is becoming affordable.

 

[8bitbytes]

The 128GB and 256GB SSDs have come down enough that the next time one of my travelers comes back with a crashed drive I am putting in an SSD.

 

[diogen]

Read the linked article: those are MLC drives - $3.5/GB at the time it was written, i.e. 6 months ago (from the table).
Even if you are OK with the specs, 20 times more per GB is not exactly affordable in my books.

 

[mike123abc]

Read the linked article: those are MLC drives - $3.5/GB at the time it was written, i.e. 6 months ago (from the table).
Even if you are OK with the specs, 20 times more per GB is not exactly affordable in my books.

I agree, hard disks have a solid future ahead for at least the next 5 years.

The price of SSDs is dropping very fast. Once a basic amount of space is available under $45 or so, SSDs will replace HDDs in most shipping systems. Computer manufacturers are mostly concerned about price, once it is cheaper to put an SSD in the system instead of an HDD, they will do it. If there was an 128GB drive at $45 right now, I bet you would see a very fast transition.