Intel Takes SATA Performance Crown With X25-E SSD

theraindog writes "We've already seen Intel's first X25-M solid-state drive blow the doors off the competition, and now there's a new X25-E Extreme model that's even faster. This latest drive reads at 250MB/s, writes at 170MB/s, and offers ten times the lifespan of its predecessor, all while retaining Intel's wicked-fast storage controller and crafty Native Command Queuing support. The Extreme isn't cheap, of course, but The Tech Report's in-depth review of the drive suggests that if you consider its cost in terms of performance, the X25-E actually represents good value for demanding multi-user environments."

Comments

Dedicated Database Storage

[Enderandrew]

This just screams dedicated database storage.

Re:Dedicated Database Storage

[Jah-Wren Ryel]

This just screams dedicated database storage.

NO, THIS JUST SCREAMS DEDICATED DATABASE STORAGE!!!

filter
fodder

Re:Dedicated Database Storage

[narcberry]

Did you even read the summary?

We've already seen Intel's first X25-M solid-state drive blow the doors of the competition

Oh, gimme more of that door knob!

Re:Dedicated Database Storage

[Eunuchswear]

So why the fuck isn't it SAS?

Re:Dedicated Database Storage

[LordSnooty]

Works brilliantly until your ads get modded down...

Re:Dedicated Database Storage

[Mattsson]

Intel probably hasn't got a controller chip for SAS yet.
So far, the big market for SSD's are in SATA, and since you can put SATA-drives in a SAS controller, they might not feel compelled to hurry through the development of a SAS flash controller.

Good price, actually.

[binaryspiral]

Considering I have a couple of HP DL380 G5s with 2.5" 72GB 15K SAS drives, each set me back about $600 (after education discount) ... the cost of this drive $738.84 with a truckload of performance to boot is a heck of a deal.

Re:Good price, actually.

[Hal_Porter]

Yup, and it uses SLC chips so it has an enormously long lifetime, around 70 years according to Intel.

Re:Good price, actually.

[nitsnipe]

As long as don't use it as swap.

Re:Good price, actually.

[cheater512]

70 years of doing what exactly?
Its entirely workload dependent.

Re:Good price, actually.

[symbolset]

SAS has taken over SCSI's role as the drive of choice to sell to customers who are willing to overpay for the cache of owning a premium product.

You could have guessed that from the full name: Serial Attached SCSI. I guess if you can't buy a new media technology from Sony, this'll work.

Re:Good price, actually.

[Wesley Felter]

15K drives exist for a reason (at least they did until now), and they're only available in SAS or FC. I suspect the SAS version is actually the cheap one.

Re:Good price, actually.

[afidel]

That's at a measly 100GB of writes per day, in a DB server it could easily see 10-100x that so between 7.5 years and 8 months.

Re:Good price, actually.

[camperdave]

Do people still use swap? I thought we just loaded the machine with lots of RAM these days.

Re:Good price, actually.

[Hal_Porter]

Actually even if you do do use it for swap, or some application that writes to out absolutely flat out the lifetime is less than Intel quotes. If I use the formula here

http://www.storagesearch.com/ssdmyths-endurance.html

2 million (write endurance) x 64G (capacity) divided by 80M bytes / sec gives the endurance limited life in seconds.

I can work out how much less.

If you substitute the figures Intel gives for write endurance (100000), capacity(80GB) and write speed(170MB/sec) you only get 1.5 years. Bear in mind that's for an application that writes flat out at 170MB/sec 24 hours a day.

The odd thing is if you compare it to the X25-M. Write endurance is 10x less at 10000, write speed is lower at 70MB/sec. There I get 0,37 years. Mind you with SLC memory being 1/3 the price you could just buy three times as much of it. That way you get 3x the storage and a lifesapce of 1 year absolute worst case. SLC actually seems like a better choice for most people.

Incidentally, this really is a worst case, hopefully no real world application can saturate write bandwidth like this.

It would also make sense to gradually decrease the write bandwidth so the drive slows down in its old age but takes longer to die. Throttling write bandwidth to 70MB/sec on the SLC drive would give a life of 3.7 years. Throttling to 70MB/sec after half the writes were used up for an average write rate of 120MB/sec would give you 2.16 years. You could imagine a sort of Zeno's throttling algorithm (50% bandwidth at 50% life, 25% bandwidth at 75% life and so on) where the write bandwidth keeps dropping so the drive slows down but never actually dies.

Re:Good price, actually.

[dougisfunny]

Since you seem to know about this, how long would a normal Disk last in that environment?

Re:Good price, actually.

[afidel]

Well I just checked and a hot disk in my SAN has done a bit over 15M 128K writes in the last 3 weeks so about 1.92TB in 21 days or close to 100GB per day. I have replaced 3 drives out of 150 in the last 2.5 years (well 5 total but 2 were precautionary from the SAN vendor when trying to troubleshoot another issue). This is a pretty lightly utilized SAN, we need it more for capacity then pure I/O. I can see a busy installation doing 10x what we do without even pushing the same hardware to its limit.

Re:Good price, actually.

[Splab]

It sure depends on your environment, my database would love these drives. We are going to be no where near 100GB writes a day for a long time, but the massive IO increase would come in handy for our reads.

Re:Good price, actually.

[Antique Geekmeister]

It is. But those HP DL:380 G5 systems were a silly design. For the same price, you can put in 6 3.5" drives in other layouts and get up to 3 times the overall storage with no perceptible speed loss. Those G5's are too much price for too little performance: if I'm going to invest in 8xSAS drives, and spend the electricity and cooling on them, I want to get some significant storage space from it.

Re:Good price, actually.

[MikeDirnt69]

Full or empty?

wicked-fast door blowing screams?

[beakerMeep]

Is it just me or have we gone full-frontal-funnyfarm with the analogies and adjectives here?

Re:wicked-fast door blowing screams?

[symbolset]

You be the judge. I would consider a factor of 80x improvement in IO/s over the best HDD, and 2x your best competitor (yourself) "wicked-fast door blowing screams" if you're looking at transaction processing for a database or other IOPS bound application. This is not the review that's overzealous about a 4% processor speed improvement. Stripe that across 5 or 10 of these bad boys and the upside potential is, um, noticable? If we can't get a little enthusiastic about that what does merit it? A flame paint job and racing stripes? A Ferrari logo? The next step up from here is RAMdisk. Yeah, it's not going to make Vista boot in 4 seconds. Is that the metric that's driving you?

Capacity is still lacking at 32GB, but obviously they could expand it now and 64GB will be available next year. Naturally if they wanted to make a 3.5" form factor they could saturate the bandwidth of the interface and stuff 320GB into a drive with no problem if they wanted to court the folks who can (and most definitely would) pay $10,000 for that premium product (HINT HINT). Obviously the price bites, but they can get it for this, so why not? Naturally for challenging environments (vibration, rotation, dropping under use, space applications, heat) it's a big win all the way around. Isn't SATA 3.0 (6Gbps) due soon?

I think I foresaw some of these improvements here some years ago. I'm glad to see them in use. If I were to look forward again I would say that it might be time to abandon the euphemism of a hard disk drive for flash storage, at least for high end devices. You can already reconfigure these chips in the above mentioned 320GB drive to saturate a PCIe 2.0 x4 link (20Gigatransfers/sec), which makes a nice attach for Infiniband DDR x4. The SATA interface allows a synthetic abstraction that is useful, but the useful part is that it's an abstraction -- you don't need to continue the cylinder/block/sector metaphor once you accept the utility of the abstraction.

Re:wicked-fast door blowing screams?

[afidel]

Hmm, at 4K DB IOPS for $719 that compares very favorably with my SAN 250K DB IOPS for ~$250K. Now for the same 7.5TB of RAID10 storage it would cost $337,050 without controllers so the SAN still wins out, but things are getting very interesting. I would expect to see drives like this make it into a high performance storage tier from SAN vendors very soon if they don't already have such an option in their lineup.

Re:wicked-fast door blowing screams?

[symbolset]

Now plug these things into your SAN -- because they plug right in -- and do the math again. 50% price premium for 80x the aggregate IOPS and 10x the bandwidth? Your SAN needs new connectors to handle the speed.

This is a slam dunk. Admit it.

Re:wicked-fast door blowing screams?

[afidel]

Depends on what I need, but it does allow for an interesting mix. Throw 300GB 10K disks for bulk storage fronted by a couple GB of write cache and these drives for things like transaction logs and you are looking at a real winning combination. Might bring down price too since you will need so many fewer spindles for the storage and should definitely bring down power consumption. The hardest part would be new installations where you don't know what you biggest users will be and what your mix will be. One of our biggest surprises is that our Lotus Notes servers are almost as rough on the SAN as the DB servers, huge numbers of IOPS and almost as much storage.

Re:wicked-fast door blowing screams?

[DXLster]

"our Lotus Notes servers are almost as rough on the SAN as the DB servers, huge numbers of IOPS and almost as much storage."

I have no idea if you fall into this category, but many, MANY Domino administrators who implement with a SAN do so in a suboptimal fashion. A Domino server should have considerable resources on local drives even if you are committed to a SAN for your primary data storage. All system NSFs should be stored on local drives, as well as your transaction log (you are using transaction logs, right?) and a dedicated indexing swap drive.

If you're running Domino 8.0.1 or higher, you should use the non-summary item compression switch on your databases, as this can improve I/O demands by as much as 30%.

The newest version of Domino is due this quarter, and includes the new attachment and object storage model that can also dramatically improve I/O, since it's possible to move things like email attachments into an alternative storage location. (Whether this is on the SAN or not is up to you.)

As an IBM design partner, I've also been pushing the Domino server team to make some further improvements in I/O for future versions of Domino. Most significantly, this would include:

1) Permitting full-text indexes to be stored in a location other than a folder adjacent to the NSF. Right now, if you're storing mail files on your SAN, you *must* store FT indexes on the SAN as well. Which is rather limiting.

2) Pulling the view index collection object out of the NSF itself. When Domino builds a view index today, the $Collection object for that view gets stored similarly (not identically) to an attachment in the NSF itself. If you look at an NSF in the Administrator client, you can right click and select "Manage Views" to see the size impact this has on the database. More importantly, this results in dramatically more I/O to the NSF itself, and increases data fragmentation. So I've been beating the drum pretty hard to get IBM to pull that object content out of the NSF. (It was a choice many years ago to store it IN the file to keep administration simple in the days when Notes servers were brought up on NetBIOS LANs in broom closets.)

If you'd like to see these I/O improvement implemented, contact your IBM representative and beat that drum with me. :-)

And if you have any questions about any of this, feel free to contact me.

Re:wicked-fast door blowing screams?

[drsmithy]

Throw 300GB 10K disks for bulk storage fronted by a couple GB of write cache and these drives for things like transaction logs and you are looking at a real winning combination.

Like this ?

One of our biggest surprises is that our Lotus Notes servers are almost as rough on the SAN as the DB servers, huge numbers of IOPS and almost as much storage.

Sounds like your Lotus servers might need some more RAM ?

Re:wicked-fast door blowing screams?

[afidel]

Nope, we generally run with a couple GB free, email is just a high I/O application. I've always tried to build my email servers like a DB server (separate raid 10's for email and logs) because modern email servers ARE database servers.

Re:wicked-fast door blowing screams?

[afidel]

We don't HAVE local drives, we use HP 1U servers with HBA's and boot from SAN. It's much cheaper to use a pizza box with SAN storage than to build a big beefy local storage server, especially when we built ours 2.5 years ago. SFF drives have increased local storage spindle counts pretty significantly though so the outlook might be different today. It's also significantly easier to manage growth on the SAN, we unfortunately lost the war on quota's so planning for growth is basically analyzing trends and growing the LUN's when needed.

Re:wicked-fast door blowing screams?

[DXLster]

You *should* lose the war on quotas. Email quotas are a terrible method of cost-control. It's trivial to implement automated server-side archiving of Domino mail instead. Setting quotas just places the time burden of message management on business users, who really ought to be doing their jobs instead of filing their email.

If you're running without any local drives whatsoever, then you're either not using Domino transaction logging (bad for reliability and write-performance) or you're putting a transaction log on a SAN (bad for I/O throughput.) System files, logs, transactions, indexing swaps -- all are easy to manage growth on because they're basically fixed-size. I've run Domino implementations for almost twenty years, and never needed more than about 10 GB for all the various system data pieces. It's more channels and heads that are the challenge, not GB count.

Forgive me if I'm breaking etiquette by saying this, but I'd love to help you tune your I/O on those servers, and improve performance. You can email me at nathan(dot)freeman(at)lotus911(dot)com

Re:wicked-fast door blowing screams?

[Hoi Polloi]

$ per GB is the deal killer here. Most of our customers balk over spending more on regular drive space when you are talking about terabyte dbs. I could only see these being good for maybe log or temp space until the price comes way down.

Re:wicked-fast door blowing screams?

[Dr. Smoove]

Hate to get off topic here but, I have to disagree with you on quotas. Quotas are a good thing(tm). "Time Burden" being key here; anything the user can do, should be done by the user. Unless you have a bazillion admins (never seen that before) then the user needs to accept some responsibility for what I consider to be trivially easy.

I really don't care who you are, from peon to CEO, part of your job is proper e-mail archival. Keep things nice and tidy with filters. Set automated pruning according to your requirements. Every so often, delete some stuff by hand, or burn it to a CD.

It's just like any other type of filing. If you were an accountant would you leave a huge pile of bills and invoices on your desk until it was so big you couldn't get to your desk?

Re:wicked-fast door blowing screams?

[srvivn21]

Hmm, at 4K DB IOPS for $719 that compares very favorably with my SAN 250K DB IOPS for ~$250K. Now for the same 7.5TB of RAID10 storage it would cost $337,050 without controllers so the SAN still wins out, but things are getting very interesting. I would expect to see drives like this make it into a high performance storage tier from SAN vendors very soon if they don't already have such an option in their lineup.

You mean something like this?

Slashdot recently covered it even.

Blowing doors of competition

[Anonymous Coward]

blow the doors of the competition

Well, that's should win over the male population....

Re:Blowing doors of competition

[Noose For A Neck]

Don't you mean the door population?

Re:Blowing doors of competition

[troll8901]

blow the doors of the competition

... and then eat them for lunch!

Re:Blowing doors of competition

[wik]

I've worked with them before. They're a real bunch of knobs.

proper comparison?

[saleenS281]

Seems to me the *target* for this drive would be the same buyer as 15k sas/scsi drives. Those are suspiciously absent from the tests...

Re:proper comparison?

[cheater512]

Do a test based on price.

$x,xxx worth of SAS/SCSI disks vs $x,xxx worth of SSD drives.

See which is faster then.
Thats the most realistic benchmark (for people without infinitely deep pockets).

Re:proper comparison?

[Anonymous Coward]

Anyone who tells you SSDs are a replacement for disks is at best talking about some niche workloads and at worst trying to sell you a line of the old BS. SSDs render 15k rpm disks obsolete all right, but not in the way you're suggesting.

To get the same capacity as you'd get out of those hot, expensive disks - which is not a lot given what you're paying for them - you'd need to spend much more, and you'll likely find that performance levels off quickly when you saturate out your HBAs and/or CPUs and/or memory bus and/or front-end connectivity. Much better to combine a few of these with slow, cool, cheap disks to maximise both performance and capacity at a lower price than the 15k disks.

Let's take an example.

Suppose you have 14 146GB 15k rpm disks. They cost you $2000 apiece, or $28000. Each one gives you 300 IOPS, for a total of 4200 (we'll ignore the costs and inefficiencies of the hardware RAID dinosaur you're probably using these with; if we didn't, you might start to feel stupid about it). So you spent about $6.67/IOPS or $14/GB, plus the power and cooling to keep those disks spinning. Not cheap. Not particularly fast. Not really great in any way.

Suppose instead that you want to replace them with these 80GB SSDs. You'll probably pay your vendor around $1400 for them (figure 60% margin like they're getting on those FC drives you've been buying from them). Now you need 26 of them to get the same capacity, costing you $36400. But you get about 12000 read IOPS each (write latency suspiciously omitted from this fluff piece, but we'll dubiously assume it's similar - it almost certainly isn't anywhere close) for a total of 312000. Too bad your HBA can do only about 140000, so you'll max out there on random reads. And if we're talking about block sizes larger than 512 bytes, latency will be higher. So you've spent $0.26/IOPS, which is great, and you've saved money on operating costs as well. But you actually spent a lot more in total - $18/GB - and woe unto you if you need more capacity; demand for storage tends to double every 12-18 months, and adding in 80GB chunks at $18/GB is going to hurt. Sure, prices will drop, but not fast enough to be competitive with the multi-TB disks we're already seeing today.

Finally, suppose instead that you buy 2 of these SSDs to act as log devices and then buy 4 1TB 7200rpm SAS disks for $350 each. You've spent $4200 and you've gotten 24000 IOPS. That's $0.18/IOPS or $0.48/GB, and you've actually spent much less in absolute terms as well. You're still spending only a tiny fraction in power and cooling of what you were spending on the original all-disk solution, and you've got twice as much total storage capacity. Best of all, you can now grow your storage in two dimensions, not just along a line fixed by the disk vendors. Need more IOPS? Add another SSD or two. Need more capacity or streaming bandwidth? Add some more rotating rust.

This approach gives you the best of all worlds, something you can't get by blindly replacing all your disks with SSDs. In other words, you get to pick the spot along the performance/cost/capacity curve that's right for your application. Using only SSDs, only slow disks, or only expensive disks doesn't do that. Upon a moment's though, this should be obvious: when your computer needs to perform better, adding DRAM is usually the best way to make that happen. When it needs to store more data, adding disks is the way to go. You don't add disks to improve performance (one hopes... if you need to do that, your storage vendor is probably taking you to strip clubs) and you don't add DRAM to increase storage capacity. This is no different. Flash occupies an intermediate spot in the memory hierarchy and has to be thought of that way. It's exciting to see the prices fall and capacities rise like they have, but I don't think a lot of people really understand yet just how SSDs are going to change things.

Re:proper comparison?

[Trogre]

s/faster/more reliable/g

Software development

[Weaselmancer]

My first thought is builds. I have to do Windows CE 5.0 builds all the time and they're almost entirely I/O bound. I've also compiled Xfree86 before at another job. It seems like the really large compiles are mostly I/O bound. The CPU doesn't peg, but the hard drive light stays lit.

Something like this would be fantastic for development. I really want one.

Re:Software development

[GigaplexNZ]

Try using a RAM disk.

Re:Software development

[cheater512]

I often do compiles (Gentoo) on a ram disk.

Linux desktop systems doesnt use anywhere near the amount of ram modern systems have so just make a tmpfs mount and the compiles fly. :)

Re:Software development

[Godji]

I'm a Gentoo user too. My CPU and hard drive are decent (Core 2 Duo 3.33 Ghz, Western Digital 500 Gb RE2). I build on the root filesystem. I've never seen an I/O bound build, it's always the CPU. What are you people talking about?

Re:Software development

[cheater512]

I mainly do it on my laptop so the HDD isnt as fast as desktops.

Still it does make a reasonable difference especially for big compiles. Try it out for yourself.
Merging is significantly faster since its copying Ram to the hard drive instead of hard drive to hard drive.

Re:Software development

[Godji]

Take your prerecorded uninformed Gentoo bashing elsewhere, please.

Re:Software development

[Godji]

I've used it on a laptop before, and I didn't notice I/O boundness either. Also, for large builds, the merge time is insignificant compared to the compilation time.

Re:Software development

[Godji]

You don't have to use it, if you don't like it. To me, it's a learning tool more thn anything else - I get to see the gory details of how certain things fit together. I expect the breakage, and generally enjoy figuring out what's wrong. I run ~amd64. And I haven't noticed much breakage lately actually.

Re:Software development

[Just Some Guy]

I often do compiles (Gentoo) on a ram disk.

So do I, but it's built-in and my system refers to it as "cache". Why not let the OS decide what to store in RAM? It's really good at that kind of stuff.

Re:Software development

[swillden]

You're trying to do the kernel's job for it, and the result is usually suboptimal. The kernel's VM system is perfectly capable of caching writes to memory, or keeping frequently-read data in RAM buffers. This is why on a properly-running system there should be little free RAM except shortly after boot.

If you really do see a measurable (not just perceived) speedup, you might want to tweak some of the kernel params a little. Specifically, increase the dirty_writeback and dirty_expire timeouts to encourage Linux to keep dirty pages in RAM longer before flushing them to disk. An easy way to do this is enable laptop mode; it'll turn both of the timeouts up to ten minutes.

In reality, I doubt you'll see a big improvement by tweaking these. I think what you're mostly seeing by using a ramdisk is the effect of pre-caching the application source in RAM when you wait while the source is copied to the ramdisk. In other words, you're shifting the time from compile time to the time it takes to copy the source into the tmpfs partition. Odds are that you could get a better result by just running something like 'find . -type f -exec cat {} \; > /dev/null' on your source tree. That way you're forcing the OS to pre-read all of the source, getting it cached in RAM.

It's also worth noting that putting stuff in a tmpfs file system does not force it to be kept in RAM, unlike the old ramdisk file system. Data in tmpfs can still be swapped out whenever the kernel feels it's useful. The only real difference is that pdflush will never try to write the tmpfs data to disk, but cranking up the timeouts accomplishes that as well.

Re:Software development

[danomac]

I am also a Gentoo user, using a QX9650 and 4GB of RAM. I did see I/O waits with my system, and then installed my root filesystem on a RAID 1+0.

The system compiles noticeably faster now - as an example openoffice 2.4.x compiles from source in about 43 minutes on my machine. Prior to that it was over an hour.

Re:Software development

[cheater512]

1.5gig of completely free memory + tempoary extensive data manipulation = perfect case for tmpfs.

Re:Software development

[Ant P.]

If you're using a filesystem with copy-on-write (btrfs, zfs, other flashy new stuff), then the merge part will be way faster on the hard disk. If those sound too scary and unstable, ext4 in 2.6.28 has the delayed-alloc thing which makes it basically behave like a tmpfs anyway until something tells the kernel to sync.
Not that it'd make any difference on my desktop - I'm stuck with a pentium 4.

It's not about speed to me

[elashish14]

I still think the biggest deterrent is lifetime. I want to buy an Aspire One, but I'm pretty disappointed at some of the things that I'll have to do with the SSD. No swapping, no journaling, no logging or timestamps. Sounds like it's still a step backwards to me. Still needs a little more time.

Re:It's not about speed to me

[elashish14]

Nevermind. "Even with 100GB of write-erase per day, it'll take more than 72 years to burn through the drive." I should RTFA. But still, much room for improvement.

Re:It's not about speed to me

[amRadioHed]

A 72 year lifespan? How much more improvement do you need? It seems like price is the only remaining hurdle for SSDs.

Re:It's not about speed to me

[Anpheus]

Why no swapping, journaling, logging or timestamps? Wear leveling is pretty standard fair for SSDs and AFAIK, at least three of those won't write a significant amount of data to the disk.

Re:It's not about speed to me

[catch23]

why do you need improvement with that? I assume you'll probably replace the drive in 10 years... 10 years ago consumer PCs used 8gig hard drives. I don't see many 8gig hard drives lying around today. Assuming you'll replace it in 10 years, you can do 700GB of write-erase per day, which means you could reflash the entire drive 20 times a day... How often do you do that on your Aspire One?

I find it strange...

[Timothy Brownawell]

This costs $22/GB, and has a write speed of 170 MB/s. A 2GB stick of DDR2-800 costs $12-$20/GB, and has a speed of 6400MB/s. So we have a case where slow storage actually costs more than much faster (but less permanent) storage. I wonder how much a couple extra batteries would cost...

Re:I find it strange...

[ltmon]

It's pretty much apples and oranges. Even with batteries (which I wouldn't trust) RAM has different characteristics in power consumption, heat output, storage density etc. By the time you address these challenges you'd have... an SSD.

Plus the SSDs get their long life from having more raw storage than advertised, and dynamically shutting down dead areas and bringing in reserve areas as it ages. Your sums would have to take into account the cost of this "hidden" storage.

As an aside the best use for these things is hands down as extended cache in a storage array. One or two SSD alongside a few terrabytes of "normal" disk managed an intelligent filesystem or storage firmware can speed the whole beast up by phenomenal amounts depending on the data usage patterns. Yet the total cost of the whole storage appliance is really not much changed in relative terms. Some of the new Sun boxes are designed to work with SSDs like this, and probably from other big storage vendors as well.

Re:I find it strange...

[cheater512]

Why would the calculations have to take in to account the hidden storage?
Thats overhead for using SSD technology. Its unnecessary for any other storage.

Adding ram purely for disk cache will increase performance many times better than using SSDs.
Cheaper too.

Re:I find it strange...

[hairyfeet]

Which brings up a question I have been wondering about: What happened to the hybrid drives? If you'll remember a few years back all the talk was of hybrids,and now that we finally have some storage that would make hybrids really cook we don't hear of them anymore. Why? Because it sounds to me like if you took 8-24Gb of this and wedded it to a nice fat SATA drive you would have an awesome laptop or gamer drive. So what happened?

Re:I find it strange...

[wisty]

Yes, but some legacy operating systems can only address 4G of RAM (including the graphics card). Also, some hardware may not be able to take more ram. I can't think of any machine where 64G of ram is very cheap.

Re:I find it strange...

[vidarh]

You can buy systems that use DRAM with battery backup that's available as a "disk". Some use standard interfaces, some use PCI cards to get higher throughput. They all have one thing in common: They are far more expensive than SSD's. In fact, one company quoted me $250,000 for a 64GB unit.

There are cheaper ones, but I don't know of any that can compete with SSD's on price - if there had been more people would've been using them as harddisk replacements. As it is, the primary market for these units are extreme high end database setups where swapping in these units is cheaper than a clustering solution or re-engineering applications.

Lousy storage density, insane price.

[SuperBanana]

Let's see...$720 for 32GB ($22/GB) versus $278 for 256GB ($1/GB.)

Keep in mind that you could buy two of those 256GB drives, mirror them, and exceed (in all likelihood) the performance of the Intel drive, and have eight times as much storage. Since reliability is pretty unproven, having them in a mirror means your ass is suitably covered.

The absolute lowest storage density (SAS doesn't come in anything less than 36GB, and 300GB is the top-end) at $22/GB, when $4/GB is the norm for SAS drives (that's a premium of 5.5x) is a big ol' cup of Fail.

Re:Lousy storage density, insane price.

[overcaffein8d]

am i missing something, or is that 128 gb, not 256?

your point remains

Re:Lousy storage density, insane price.

[tfranzese]

Except those same drives don't exactly compare due to a poor implementation of the hardware or the write/cleaning algorithm in the JMicron controller many (all?) of those are using. The capacity and price are tempting, but the write latency especially during random accesses is beyond awful. Unless of course they were able to update the firmware on those chips to address the issue since this article was published: http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3403&p=7

Re:Lousy storage density, insane price.

[0100010001010011]

You got my hopes up. That's 128GB.

Re:Lousy storage density, insane price.

[atamido]

Most of the JMicron controlled SSDs can not have their firmware updated. But I believe the controller chip itself is rather broken, so there is little a firmware update could do anyway.

NCQ on an SSD?

[mseidl]

OK, someone explain something. How does NCQ work on a SSD drive? It isn't spinning, so what's the point?

NCQ to my understanding was stolen from SCSI and thrown onto SATA drives.

So, the data on the drive is in the order "2 3 1 4" and you request it in "1 2 3 4" w/o NCQ it would take 4 revolutions to pick up all 4 requests. With NCQ it would take 1 revolution as it would pick up the data as it was in the drive.

Re:NCQ on an SSD?

[FunkyRider]

Opening a bank/row to allow the memory cells to be read/written takes time, too. In RAM terms, that's called CAS Latency. This is what re-ordering helps to reduce: switching banks/rows

Re:NCQ on an SSD?

[Rayban]

NCQ gives the SSD something to do while the host is figuring out what to write or read next. Normally it's used to allow the host to fire and forget 32 commands. In this case, you queue up a bunch of stuff, then figure out what to queue next.

SSDs are so much faster that the host is generally not keeping up with it.

Re:NCQ on an SSD?

[Clarious]

I have read somewhere that NCQ here is to help the OS follow the speed of the HDD, not for the HDD to follow the speed of the OS.

Re:NCQ on an SSD?

[Bacon Bits]

Even if it's solid state, there's still a physical layer, and still a logical-to-physical abstraction that an IDE disk must perform. (Slashdot pedantics will please note that here an IDE disk means a disk with an integrated electronic controller, not just a drive with an ATA interface. If you've never had to know the true physical geometry -- the number of cylinders, heads, and sectors in a disk (CHS)-- to tell your PC's BIOS or OS, you've never used a non-IDE disk. Most BIOS systems were faking CHS numbers by the time EIDE hit in 1994 which eliminated CHS in favor of LBA.)

Flash drives use NAND flash memory, which uses pages of up to about 4 KB. For the most part, you can only access a single page at a time. Additionally, sequential access within a page is almost always faster than random access. Giving the disk's integrated controller a list of values means that it can examine the queue intelligently and can perform paging operations more intelligently.

Re:NCQ on an SSD?

[afidel]

Basically allowing the drive to have a queue of requests and responses outstanding. Your host will often be bursty and want to send a bunch of commands at once which even an SSD won't be able to service, but it can service them much faster than the host can issue, receive, issue etc. Think of it as turning on Async I/O at the hardware level.

Re:NCQ on an SSD?

[shentino]

If only the RIAA understood that...

it costs more per gb than ram!

[phr1]

PC-6400 ram is around 15 dollars a GB now, and the 6400 stands for MB/sec, i.e. ram is over 20x faster than this flash drive and has no write wear issues or slowness of random writing. The only thing wrong with it is volatility, but in an enterprise environment you can use a UPS and/or maintain a serial update log on a fast hard disk or RAID (since the log is serial, the flash drive's ultrafast seek advantage doesn't apply). There is just a narrow window where this $21/gb 32gb flash drive is really cost effective.

Re:it costs more per gb than ram!

[dark_requiem]

Products that use RAM as the storage media have been around for years. They're exactly what you're describing: a few standard DDR DIMMS and a battery on a PCI card, usually. However, no one in an enterprise environment would actually trust data to such a device, and they never really took off. Home users don't generally have the power and data backup capacity to safely use such a device (and not even the most hardened masochist wants to reinstall or restore everything whenever a breaker goes), and enterprise users can't tolerate the risk level. Sure, you can have backup power, but the risk of losing data and downtime restoring it just isn't tolerable in most enterprise environments.

Re:it costs more per gb than ram!

[PacketShaper]

True, but RAM does lack that one little feature most "storage" needs... persistence.

Re:it costs more per gb than ram!

[Splab]

Actually, RAM-SAN has addressed these issues, only thing remaining is the price tag. Think they come in at around 430.000 IOPS, got full drive backup + batteries in a single "box". Would love to get one of those for my database.

Re:it costs more per gb than ram!

[Splab]

Ok, got my facts wrong. 3.2 Million IOPS, 24GB/s sustainable random data access.

And a linky:
http://www.superssd.com/products/tera-ramsan/

Re:it costs more per gb than ram!

[Kent Recal]

Yup, violin basically sells the same thing - in a yellow box, too.
The problems with these boxes, until recently, have been price and (obviously) persistence.

Since data is never really persisted you'd only buy them in addition to a traditional SAN (or SSD nowadays), not as a replacement.
When you have the dough you can do interesting things with them, though. I know a company that does most of their transaction processing on violins (financial sector, sick throughput) and uses spindles in a write-behind fashion. Technically very interesting but will imho always remain a niche market.

Re:it costs more per gb than ram!

[Splab]

You should read up on Tera Ram-San - the data is indeed persistent, it even comes with internal backup making sure it can dump all data to its internal discs before shutting down.

Re:it costs more per gb than ram!

[Kent Recal]

...unless it fails, I suppose?

Admittedly I haven't read up on them but most people I know wouldn't be comfortable with such a "persist on shutdown" option - because the interesting scenario is when the box doesn't get a chance to shutdown.

Re:it costs more per gb than ram!

[Splab]

Well for the box not to get a chance to shutdown would require something very terminally happening, and in that case you are going to have to run down and grab your backups anyways.

Also as I recall the system will periodically flush everything to discs so it doesn't have to make the full write in case of emergencies.

Re:it costs more per gb than ram!

[Kent Recal]

Flushing to disk imho can not work so well when the box is loaded with more IOPS than the spindles can handle. It's the same problem that every database server is having (RAM vs WAL vs tablespace).

And about the shutdown issue: When you're talking about a scale where such RAM-based SANs get serious consideration then unknown risk factors like "terminal happenings" can usually not be tolerated.
Building in the enough safe-guards to make such a failure sufficiently unlikely is usually more expensive than building a redundant setup in first place. Hence in most setups you'll probably see a traditional SAN with permanent storage in the mix, rather than a "RAID10 over RAM-SANs" without other fallbacks.

I haven't worked with RAM-SANs first hand but I know that even in the traditional SAN world most people will, for mission critical applications, buy at least two heads, even when each has plenty of redundance (including an UPS) built in.

Re:it costs more per gb than ram!

[atamido]

Flushing to disk imho can not work so well when the box is loaded with more IOPS than the spindles can handle.

Most high IOPS systems are high because of a high number of random reads. The spinning disk would only need linear writes, which should be pretty easy to sustain maximum transfer rates. That Tera-RamSan appears to use 128GB modules, so each would include its only spinning backup disk. Modern disks can easily sustain 100MB/s. A full dump would take 20 minutes.

I suspect they would at least implement a basic paging system where the RAM would have a 1:1 mapping on the harddrive. Then using block sizes (100MB or 10MB maybe) mark whenever a block has had data changed and is no longer current. Then take the changed block that has not been modified for the longest time and dump it to disk. Even for most large databases most data is written onto the end (or log files) so there are typically large contiguous sections that aren't written to often. And note that doing a mem copy for 100MB and then writing to disk is not going to be a blocking action.

It's possible that (depending on the load) a shutdown initiated flush of changed blocks wouldn't take very long. And more efficient algorithms could really strip that down.

Question for slashdotters:

[nitsnipe]

What happens when the read-write cycles on this run out?

Re:Question for slashdotters:

[WhatAmIDoingHere]

Well, it'll be 2080, so hopefully you would have replaced the drive sometime in the 2020 - 2030s.

Re:Question for slashdotters:

[cheater512]

It becomes a very fast cdrom - read only.

Re:Question for slashdotters:

[MoogMan]

A *much* better failure model than HDDs forgetting about data on dodgy sectors etc.

Weak test system

[dark_requiem]

I would have liked to have seen them test this drive in a much more powerful system. I mean, a P4 with 1GB RAM, and a fairly dated chipset (955x) as the SATA controller? No one is going to put a drive like this in a system that old. I'd guess that we might see different results on a more powerful system. At some point in those tests, other components of this fairly slow (by today's standards) machine. Throw some serious power behind it, and you can be sure that you're not bottlenecked, and the full power of the drive shows. Can't say for sure if this is actually the case, as I don't have a drive to test, but it's a definite possibility. Hopefully someone else does a similar review with a more powerful testbed.

NCQ?

[melted]

Why the heck does a drive that has uniform, low latency random access would even NEED NCQ? NCQ was designed to optimize the seek order in mechanical drives with heads.

Re:NCQ?

[Anonymous Coward]

From the article: "The storage controller is an Intel design that's particularly crafty, supporting not only SMART monitoring, but also Native Command Queuing (NCQ). NCQ was originally designed to compensate for the rotational latency inherent to mechanical hard drives, but here it's being used in reverse, because Intel says its SSDs are so fast that they actually encounter latency in the host system. It takes a little time (time is of course relative when you're talking about an SSD whose access latency is measured in microseconds) between when a system completes a request and the next one is issued. NCQ is used to queue up to 32 requests to keep the X25-E busy during any downtime between requests."

Re:NCQ?

[Anonymous Coward]

read here: http://techreport.com/discussions.x/15374

Since we had her cornered, we took the opportunity to quiz Huffman about a few other matters. One of those was the interaction of Native Command Queuing and SSDs. We're familiar with NCQ as a means of dealing with the seek and rotational latency inherent in mechanical hard drives, but wondered what need there was for NCQ with SSDs. (Intel's just-announced SSDs have NCQ listed prominently among their specifications.) She said that in the case of SSDs, NCQ has the primary benefit of hiding latency in the host system. If you look at a bus trace, said Huffman, there's quite a bit of time between the completion of a command and the issuance of another one. Queuing up commands can keep the SSD more fully occupied.

Re:NCQ?

[Kent Recal]

In soviet russia...

Not a good price, actually.

[symbolset]

And neither one is as reliable or has the IOPS of three standard SATA drives that cost 50% of the price for 10x the storage.

Math. It's a wonderful thing. Use it with your salesman.

Re:Not a good price, actually.

[Splab]

So is ignorance apparently.

SAS is duplex, SATA isn't. I'll take one SAS drive over 3 SATA drives any time when it comes to performance.

Re:Not a good price, actually.

[drsmithy]

Math. It's a wonderful thing. Use it with your salesman.

Have you used your "math" to figure out how much more tripling rack space requirement and doubling the power consumption will cost ?

Re:Not a good price, actually.

[symbolset]

There are actually a few compelling use cases... Short of space is one of them. Server consolidation has freed up a lot of rack space lately, though, so most people have the space.

And we're talking about drives that burn under one watt running full out. How many do your SAS 15K RPM drives burn?

Re:Not a good price, actually.

[symbolset]

Here's where I would go with a useful link. "Duplex" doesn't necessarily mean what you think it means in this context. The use of this term bare is misleading, as perhaps the marketing person who invented the meme intended to be.

"FYI, SAS full duplex means that one channel can be used for data traffic and the other channel can be simultaneously used for command traffic. Both channels cannot be simultaneously used for data. So when Mr Batty says 6Gb/s is available and that's 4x SATA I, he is technically correct, but end users will not see 4x performance."

If you can't sell on the features, it's ok for some people to make stuff up when they're selling. But not us, here, ok? Let's be honest with one another here around the water cooler.

Re:Not a good price, actually.

[drsmithy]

And we're talking about drives that burn under one watt running full out. How many do your SAS 15K RPM drives burn?

Which SATA drives use less than a watt *at all*, let alone "full out" ?

SLC too!

[Trogre]

Ooh, now that could be a dealmaker in the server room. With RAID-5 reaching its limits for magnetic media, a rack of these could be a viable replacement.

Of course a server room has different priorities to the average gamer:

1. Reliability
2. Reliability
3. Capacity
4. Price
5. Speed

Re:SLC too!

[troll8901]

With RAID-5 reaching its limits for magnetic media, a rack of these could be a viable replacement.

* bracing self for long discussion on RAID levels, file systems, and a certain Unix OS *

Re:SLC too!

[vidarh]

RAID5 is not for performance but for reliability, and it's only "reaching its limits" for large capacity systems where the time to rebuild the RAID on drive failure is getting close to the point where the risk of a second drive failure during the rebuild is becoming an issue.

These drives are intended for high performance setups.

It's not RAID vs. these drives - nothing prevent you from using them in RAID setups, and most people using them in servers probably will.

Performance that pebbles

[cstec]

Allright the price, well, that's to be expected. But the performance rocks! Just ... itty, bitty rocks.

32Gig? 32Gig?!? Come on, there is probably have more than 32Gig on this drive just in Vista system restore points.

Re:Performance that pebbles

[cstec]

E'glish speak good me now! Neanderthal clone bad!

Re:Fusion IO still whips it.

[isorox]

Meh, still not even half as fast as the Fusion IO-Drive. Of course, those cost $3,000+ and run solely on a PCIe 4x ...

We're trialing one now, not interested in production until there's solaris drivers though.

blow the doors of the competition?

[tyrione]

I believe you meant to say, ``blow the doors off the competition,'' but let's just say the "cost" of those drives ``blows chunks.''

Extreme for enterprise

Solid-state drives use either single-level or multi-level cell flash memory. The former stores one bit per memory cell (a value of 0 or 1) while the latter is capable of storing two bits per cell (with possible values of 00, 01, 10, and 11). Obviously, MLC flash has a significant advantage on the storage density front. However, that advantage comes at the cost of write speeds, which are typically much slower than reads. Intel's MLC-based X25-M, for example, is capable of reading at up to 250MB/s, but its sustained write speed tops out at only 70MB/s. Single-level cell memory doesn't suffer such a great disparity between read and write speeds, as evidenced by the X25-E Extreme, which reads at up to 250MB/s and writes at up to 170MB/s.

Of course, the more balanced transfer rates offered by SLC memory don't come cheap. The X25-M 80GB is currently selling for $621 online, which works out to a seemingly exorbitant $7.76 per gigabyte. But that's nothing compared to the cost of the X25-E Extreme 32GB, which at $719 online, rings in at an even steeper $22.47 per gigabyte. Solid-state storage isn't cheap, and single-level cell implementations are about as expensive as SSDs get.

I hear the US Government knows how to piss money down the drain. I'm bettin' they think this price/Gb is just f'n dreamy!

Blowing The Doors off

[GrahamCox]

It's the late 60s and a groupie is invited backstage after a particularly mega concert featuring all the great bands of the day. After a while, a bit of pot has been smoked, some tabs dropped, and plenty of booze swigged, and so things start to swing. The groupie first goes down on Ray Manzarek, then Jim Morrison, and finally Rob Krieger and John Densmore get theirs. Groupie's not done though, and is just getting started on Jimi Hendrix when Michael Caine bursts in, and shouts...

"Oi! You're only supposed to blow the bloody Doors off!"

Re:Blowing The Doors off

[GrahamCox]

And for those (probably non-British folk) who didn't get the reference: You tube clip

Does the performance degrade with use?

[Ed Avis]

The earlier model of Intel SSD had some serious performance degradation after a few hours of heavy use. (Article in French, but it says that after a ten minute torture with IOmeter writing small blocks to the drive, and even after waiting an hour for the drive to 'recover', performance drops by about 70%.) I wonder if they have fixed this bug with the new model?

Re:Does the performance degrade with use?

[atamido]

They state specifically that the controller arranges data for typical data access patterns, and that synthetic benchmarks writing randomly all over the drive screw with the optimizations they've done. I'd like to see tests done with real server software to see if the performance degrades significantly.

Re:Does the performance degrade with use?

[Ed Avis]

The synthetic benchmark running slowly is not the problem. What is worrying is that you can run the benchmark once and it somehow crufts up the drive so that all future operations run much more slowly. The only way to get back the original performance is a low-level reformat. That is the bug that Intel needs to fix.

Comment removed

[account_deleted]

Comment removed based on user account deletion

Performance crown and new-school math

[bconway]

We've already seen Intel's first X25-M solid-state drive blow the doors of the competition, and now there's a new X25-E Extreme model that's even faster. This latest drive reads at 250MB/s, writes at 170MB/s

Yet, 5 articles down the Slashdot homepage, options depending:

Samsung said it's now mass producing a 256GB solid state disk that it says has sequential read/write rates of 220MB/sec and 200/MBsec, respectively.

I'm pretty sure the improved write speeds is the part that people are interested in with SSDs these days.

Non-volatile RAM disk ?

[zrq]

From the techreport article :

NCQ was originally designed to compensate for the rotational latency inherent to mechanical hard drives, but here it's being used in reverse, because Intel says its SSDs are so fast that they actually encounter latency in the host system.

Is it time to look at connecting these chips direct to the motherboard ? Avoiding the added complexity of driving what is essentially a block of memory via a serial interface designed to control spinning discs. If the SLC memory chips were mapped into the main memory address space, it should be possible to make them look like a 32G or 64G (NV)RAM drive on a Unix/Linux system. Mount '/' and '/boot' on the (NV)RAM drive and install the OS on it. Presto - very fast boot and load times. You can still use traditional spinning disc(s) for large data, mounted as separate data partitions.

It would need some thought as to which parts of the filesystem went on spinning disc and which parts went on the (NV)RAM partition. But that is why Unix/Linux has all of the tools for mounting different parts of the filesystem on different partitions. Back in the olden days, most systems had a combination of small fast(ish) discs and big(ish) slow discs, and tweaking fstab to mount different parts of the filesystem on different discs was a standard part of the install process. Most desktop systems now have one huge disc, and the standard Linux install dumps everything on one big '/' partition, but all the tools for optimizing the partition layout are still there.

How about an ultra quiet desktop workstation with no moving parts, the OS installed on (NV)RAM disc, and user data dragged across the network from a fileserver (e.g NFS mounted /home).

Comment removed

[account_deleted]

Comment removed based on user account deletion

ZFS should like these

[Fweeky]

ZFS in recent Solaris Nevada and FreeBSD CURRENT supports a feature called L2ARC; Level-2 Advanced Replacement Cache. This is basically the ZFS filesystem/metadata cache, backed by so-called cache devices.

So, you can get your 32GB SSD, shove it in front of your n-TB ZFS array, and it'll use it to help accelerate random reads. 32GB of storage is a bit feeble, but 32GB of cache.. that's rather compelling, especially if your storage is otherwise backed by cheap and cheerful 7200RPM disks.

Fusion IOdrive is faster

[dh003i]

The Fusion IOdrive is faster...about 510 Mb/s according to dvnation. At $30/GB, that's not bad. Granted, the Intel one is $22/GB...but it has about twice the performance; and it's only priced about 1.4x the price of the Intel ssd.

Memoright / Mitron?

[foxalopex]

Hmm odd that seemed to have missed out on the leading enterprise SSDs. I think Tom's Hardware reviewed them a while back. Samsung SSDs from what I remember were as cost effective as they get but generally were slower than either Memoright or Mitron?

"anything the user can do..."

[DXLster]

I guess we'll agree to disagree. I'd submit that anything an administrator can automate (eg: all email older than 90 days will be rolled over to an archive system with a lower SLA) should be automated. Anything a computer can do instead of a user should be done by the computer.

If I'm an accountant, the whole reason I use a computer is so there is no pile of bills and invoices on my desk. If, when I'm done working on a client's books, the computer knows where to tuck them away for me automatically, then that's good system design.

Re:Dedicated Database Storage & MORE... apk

[Celandro]

Saying the i-RAM is in a league of its own is no longer accurate. The X25-E is directly comparable to the i-RAM.

http://techreport.com/articles.x/15931/9

The i-RAM is only twice as fast as the new flash based drive. That a storage device is anywhere close to ram in speed is a huge advance in the industry.

As for even read/write speeds, and other old flash drive problems, the new Intel drives are very well designed and do not have the same type of issues.

Re:Stupid Question!

[Eunuchswear]

3) SCSI adds 1 about 1ms of latency due to processing in the host bus
adapter and the device.

Why would a SAS adapter/drive process SCSI commands slower than a SATA adapter/drive?

Moving on, SAS adapters cost more. Whereas SATA is free with most modern motherboards.

I can't think of the last time I connected an important device to a motherboard controller. All my disks are in external boxes.

The one in the article

[symbolset]

Here is the specific page of TFA about the objects under discussion. 0.62W at idle.

Now about full out... I don't have a citation but if you get more than 1.4, let me know. That would be... interesting.

On the upside for SAS, these SATA drives plug right into your SAS array, so they've got that going for them. They don't make a SAS version of this drive yet but SAS controllers are compatible with SATA drives (although the reverse is not true). But of course you knew that. Yes, I know an SSD is technically not a "drive". If that's where you were headed don't bother. It would be tedious and pedantic.

Re:The one in the article

[symbolset]

I meant 1.8 of course. Seeking tests in TFA give 1.6. Typical SAS 15krpm platter drives are almost 5x that.

With SATA SSD, thermals and consumption are not issues. IOPS and throughput are a clean win, and the SAS interface is no advantage because SAS spinning drives can't deliver the throughput, latency or IOPS of SSDs because they are not solid state. Price per GB is comparable with high-end SAS. The last critical measurement is pure volume per drive and GB/$ and neither SAS nor SSD can compete with SATA 1.5TB drives in that arena.

When we talk about the increased performance of rotational speed, why do we ignore the benefits of areal density? Why? A 1.5 TB drive has something like 25x the data flying under its head at 7200 RPM as a 72GB 15kRPM SAS drive does. Why in the world is this factor not important?

Re:The one in the article

[drsmithy]

Yes, I know an SSD is technically not a "drive". If that's where you were headed don't bother. It would be tedious and pedantic.

I'm "headed" that way because your original comment was:

And neither one is as reliable or has the IOPS of three standard SATA drives that cost 50% of the price for 10x the storage.

Don't move the goalposts. You were originally criticising SAS because you thought "standard SATA drives" needing three times as much space and about twice as much power were a better deal.

[...] and the SAS interface is no advantage [...]

Actually it is, because SAS devices nearly always support multipath IO as standard. On the SATA side, however, *very* few SATA backplanes support multiple controllers (and even fewer drives - off-the-shelf drives will pretty much always need a multiplexer).

Price per GB is comparable with high-end SAS.

A 73G, 15k RPM SAS drive costs about US$140. The 32G SSD from TFA costs about US$720. Per GB, it's an order of magnitude more expensive.

When we talk about the increased performance of rotational speed, why do we ignore the benefits of areal density? Why? A 1.5 TB drive has something like 25x the data flying under its head at 7200 RPM as a 72GB 15kRPM SAS drive does. Why in the world is this factor not important?

Firstly, because the difference is not that high. High-performance SAS (and SATA, for that matter) drives do not have full-size platters. Secondly, because areal density has a relatively low impact on IOPS, which is the number most people looking at high-performance disk are interested in.

Yes, if you want loads of disk space, or high streaming performance, a high-density SATA disk is a reasonable solution. However, when most people want "performance", they want high IOPS with random access patterns, which is something "standard SATA drives" don't deliver without significantly higher spindle counts. SSD does, but the per GB costs are vastly higher than 10k or 15k SAS drives.

The overriding point here is that all the different storage technologies have their place, depending on your requirements.

Re:The one in the article

[symbolset]

You make some good points. I think that's a thorough coverage of the topic.