View Single Post
Old 04-04-2013, 02:44 AM
anthony.k11 anthony.k11 is offline
Join Date: Apr 2013
Location: U.S
Posts: 2
Default Re: How About Solid State?

Originally Posted by Darryl Ramm View Post
"Wear leveling kills write performance" .... Sigh, where do I start. "kills" is just competently misleading hyperbole. Wear leveling does affect write performance, but just saying that is not really helpful in any practical sense. Yes flash SSD drives do wear leveling, they have to, and that has some overhead, as does other things the drives do. But so what? All that matters is the actual sustained sequential and random write and read performances that modern SSD drives can achieve.

It would be nice if instead of scaremongering about SSD like this, the focus was on discussing the actual technical capability/performance of current SSDs, and how to best configure and use those drives. Modern high-end consumer and enterprise class SSDs are easily capable of meeting typical and high-end audio needs, and more importantly can exceed the practical performance of single high performance HDD. Especially, they can meet high sequential performance and also handle additional random IO perturbation as we can have on audio drives. Unfortunately we seem to keep seeing very out of date information and misperceptions about SSD drives circulating, and that seems to be happening here.

And while I am an advocate of using (high-performance) SSD drives with Pro Tools, and believe they can be excellent for system/boot, sample and audio drives I am a little surprised by Avid's simplistic position that all SSD drives are supported with Pro Tools 10. Performance, especially sustained sequential and random write performance of SSDs varies widely, it can *suck* on some older drives and on many drives if improperly configured, and I just don't see Avid talking about correct SSD configuration/use anywhere.

I'm just not getting the relevance of referencing the EBU or Micron papers here. The EBU technical paper mentioned here is from 2011 and while it is fairly well written and insightful, it is largely a historical background piece as it (a) was written in 2011, and (b) discusses fairly irrelevant drive performance from a ~2009 model drive and (c) as that paper points out the the system used was seriously misconfigured including not having TRIM support enabled amongst a long list of other issues. The Micron paper, just describes how NAND Flash SSD writes work, nothing really pro or con using SSD comes out of that, its just basic operation stuff, although unfortunately the article fails to mention TRIM and its typical importance on sustained write performance.

Current model high-end consumer-grade SSDs (what most Pro Tools users are likely to consider) are significantly better than they were way back in 2009. There has especially been huge improvements in on-board controller technology, the firmware they run and the resulting garbage collection performance (that allows much higher sustained write performance than older drives). These modern SSDs support very high write rates, and for typical models especially with TRIM from the operating system offer can *sustain* many hundreds of MB/seconds sequential writes. Slight over-provisioning (like typically formatting the drive and leaving 10%-20% unallocated space) also helps improve garbage collection efficiency and so helps sustained write rates. And that over provisioning provides a pool of flash pages to replace any that wear out (the drives by default come over- provisioned for this purpose, but it often helps to increase the overprovisioning if you know the drive is going to be written to heavily). In Windows land, you need Windows 7 to support TRIM (so typically forget trying to get high sustained write performance with SSDs with Windows XP or Vista, it's just a non-starter). OS X 10.6.8 added TRIM support, although that often require enabling TRIM with the third party Trim Enabler Tools

Its also worth remembering just how *underwhelming* many of our audio workloads are for a modern SSD. Data just does not get written to these drives very fast, not compared to what modern SSD drives can do. e.g. to hand wave some crude numbers 128 tracks @ 32 bits @ 96kHz = 48 MB/sec. (actual Pro Tools numbers will add some overhead above this, but this gives a useful start). As we'll see 48 MB/sec is *nothing*...

My current favorite consumer grade MLC SSD is the Samsung 840 Pro. And probalby the best review of that drive is by Anandtech, and that review is pretty educational to read. See Notice things like the 413 MB/sec *sustained* sequential write speed. And that is sustained as the benchmark fills the entire drive... Again remember those sequential Pro Tools writes from that 128 track 96 KHz session is only generating 48 MB/sec. Now the actual write patterns from Pro Tools is a little different but the raw performance of these drives is staggering.and as dramatically pointed out in the TRIM/No TRIM comparison tests in the review, that requires TRIM support to be properly enabled or the performance tanks (the benchmark warm-up code already screws with the garbage collection/dirties up the SSD drive free page lists that the benchmark results show it starting at awful write performance and slowly staggering back to life), no surprises there, working exactly as expected. message si you turn TRIM on with these drives and that does not happen.

Now to hand wave about drive life.... Using that rough estimate: 128 tracks @ 32 bit @ 96kHz writing to a 512 GB SSD it would take ~ 2.5 hours to fill the entire disk (I've taken out a 20% over-provisioning allowance).

For those 128 tracks, if we have an SSD with say (guessing) 10,000 write cycle, then the most naive math says the the disk would last for ~ 25,000 hours or ~3 years of continuous write use. Find me anybody who has 128 tracks running continuously like this and I'll find you somebody happy as a clam to replace those drives every few years. The reality is for most Pro Tools users they just won't get near to wearing out the SSD, which are more likely to be replaced because they become interface-standard or performance obsolete.

And now to be fair this is a simplistic calculation, an actual calculation needs to understand lots of internals (that the drive manufacturers do not make public), including details of ECC algorithms, actual wear patters seen and expected actual page life, vendor Flash burn in and chip selection trick, etc.. I'd love vendors to provide a lot more detail of these geeky internal things but I suspect they are all paranoid, and especially do not want to disclose things to competitors. But I'd like the see the industry as a whole improve on specing/quoting drive wear rates and things like TDW (total data written) specs. But here is the thing, folks are using SSDs in lots of much demanding IO applications, including write intensive apps, and high quality drives just are not falling over a lot. Vendors like Apple are pushing SSDs in lots of their laptops and other computers. There are many millions of personal computers in the field running SSDs in all sorts of workloads, and you just don't see horror stories of them falling over all the time.

Return to one of my favorite review sites AnandTech for a discussion on flash wear issues (summary, with MCL its a mostly a non-issue again, fold in your own write speed estimates). But go read the details. and this is for an older Intel SSD with a SandForce controller (not my favorite) but its all about the discussion of flash wear cycles that matters.

I run SSDs for everything with Pro Tools, nothing particularity demanding in my own use, and I frequently push the drive writes significantly harder doing software development and other things unrelated to audio. Other folks on DUC run Pro Tools audio on SSDs. I've not heard reports of them having problems on any of these current generation SSDs. If they were so problematic I would have expected that we would be starting to see folks complain....

At the very high-end of that spectrum my day-job puts me close to some flash related hardware companies and software companies that develop specialized database application that write at very high rates to SSDs. These are orders and orders of magnitude more demanding than any audio application. For the class of Enterprise/Internet problem they are trying to solve they'll typically use SLC flash drives, sometimes SATA and sometimes PCIe based. The folks are running many tens of TB of SSD drives on single systems in write intensive applications that have proven to be very reliable (when properly engineered) and they have gotten to a point where drive failure is really just not an issue, and actual reliability in the field over a couple of years of use is significantly better than the large number of 15k SAS disk drives that would be needed to deliver equivalent performance. That does not prove anything for audio applciation and consumer grade MLC devices, except to point out that high-end enterprise and Internet users were also doubting a few years ago that even with SLC wear rates that this technology would work for their needs. Its gone very quiet about this in Enterprise and Internal storage land recently.

There have been a lot of great things happening with SSDs over the last few years, SATA MLC SSDs for consumer applications, SATA and SAS SLC SSDs for enterprise applications, PCIe-flash cards mostly for enterprise applications and some high-end power user apps. And to me the most interesting things happening right now are PCIe-SSD "express flash" drives, 2.5" form factor drives with multiple PCIe edge connectors, that deliver PCIe flash card type performance (potentially several times higher IO metrics than a SATA based SSD) with all the easy hotplug and field serviceability of conventional 2.5" SATA enterprise drives (these SSDs just look superficially like a 2.5" SAS or SATA drive, they directly attach to PCIe lanes not to SAS or SATA). Early innovators especially Dell and Micron are packaging some incredibly powerful server systems based on this technology. e.g. see and That is likely what the future of any network NAS and SAN storage "bricks" look like. and hopefully well see the 2.5" PCIe-SSD form factor appearing in high-end workstations as well. And wrapping it all up in a bow, there are companies like Violin Memory, IBM/Texas Memory Systems and PureStorage that are packaging up Flash technology as extremely high performance NAS/SAN solutions and often doing impressive end-runs around traditional storage infrastructure providers that often seem chained to their HDD legacy technology. So expect all your NAS/SAN storage in larger sites to go that way as well.

BTW here is another (few month old) review of leading SSD drives is from PC World.

As a technology sector, the spinning HDD manufactures are on their last legs, HDD is still technology that will replace tape for quite some time, and will be used where huge data sets make SSD too expensive (online video archiving/libraries comes to mind). But HDD technology had its day and there is likely nothing more coming that can save those manufactures unable to move to SSD production. Proper spec HDDs will continue to work great as audio recording drives for many users, and like all things if there is no reason for you to change, then don't. But many user will see benefits from SSD including... increased performance, increased reliability, less sensitivity to mechanical motion/shock, lower noise, lower power consumption, etc. All HDD user will eventually get replaced by SSD technology over time. And more change is already on the distant horizon, current SSD is NAND flash, over time other technologies may come along (e.g. MRAM chips are just starting to appear from Everspin Technologies, they are very interesting, lack some of the current "challenges" of nand flash, like slow page erases, and any wear at all, but still have a very very long way to go to directly replace nand flash).

Anyhow lets please have less rehashing old and out of date information and more focusing on what is actually achievable with today's SSD products. Maybe other folks using SSDs with Pro Tools (as audio or other drives) can mention their experiences.

Darryl Ramm, let me share my SF success story with you. I have used Intel 520 SSD based on SF controller & feel more than happy. This has Data compression built into it & has good benefits which makes it stand apart from other SSD competitors. This means that this built in data compression holds good for all SF based SSDs. I think this is quite innovative approach by them. If you have a look at this white paper(link towards end), Intel has quite clearly indicated that 75% of the file types can be compressed 60% or more. Also, compression decreases number of writes to the NAND & hence endurance is enhanced. Ref:
Reply With Quote