Intel Intros 310 Series Mini SSDs

crookedvulture writes "Intel has added a couple of tiny 310 Series solid-state drives to its storage lineup. Measuring just 51 x 30 x 5.8mm, the mini-SATA SSDs are about a tenth the size of a standard notebook hard drive. Impressively, their performance ratings track with full-sized SSDs. Intel is pushing the 310 Series as a solution for dual-drive notebooks that combine solid-state and mechanical storage to give users the best of both worlds. Next-gen notebooks just got a little more interesting."

Comments

Drat

[DurendalMac]

I was excited as these appear to be Mini PCIe cards, but then I was disappoint as it looks like it's a SATA connector that shares the form factor. It's not entirely clear, though.

Re:Drat

[fuzzyfuzzyfungus]

It isn't a vice exclusive to Intel; but that is indeed what you are seeing.

For reasons that I can only imagine had something to do with "somebody pinching pennies until their pecuniary ichor flows", the trend somehow started of using the mini-PCIe connector, without so much as the decency of different keying or anything, to handle what are, electrically, SATA signal lines plus power. There would be nothing wrong with this if these things were actually storage-oriented mini-PCIe cards(like the HDD PCI cards of yore, with a controller chip+flash, capable of acting like a normal PCIe device; or if they were just using some 'sub-mini SATA' connector; but using a straight mini-PCIe connector for something electrically and logically completely different is plain hostile.

I get this sense that users aren't really supposed to touch these things, or the innards of the devices in which they will end up, or such a confusing and potentially damaging connector misuse would likely have not taken place...

Re:Drat

[dreamchaser]

Why is SATA a disappointment?

Because slightly older laptops might have an emtpy Mini PCIe slot but not an extra SATA connector? To me it's not a dissapointment but perhaps to the poster you replied to it is.

Re:Drat

[Rockoon]

SATA 1.0 (1.5 Gb/s) can't keep up with any modern SSD

SATA 2.0 (3.0 Gb/s) is currently keeping the industry down.

SATA 3.0 (6.0 Gb/s) isnt widely adopted yet, but even when its finally popular enough that too will just keep the industry down.

SATA-IO should be ashamed of itself for implementing 3.0 with such bullshit specs given the obvious reality of the situation.

Thats why many people want PCIe to become a standard interface for SSD's. That wont happen until low cost/capacity SSD's use it.

Re:Drat

[fuzzyfuzzyfungus]

Until you hit the really high end(where SATA is a bottleneck), there isn't much wrong with SATA. It's more the fact that mini-PCIe slots, sometimes several, are downright standard in notebooks and similar small devices, while these strange, hybrid 'electrically SATA; but mini-PCIe connector' things are not. an SATA device isn't going to do anything useful plugged in to a conventional mini-PCIe slot, and it will require a mechanical adapter to connect to any reasonably normal SATA connector.

For the moment, unless these things really take off, this particular combination of form factor and bus type screams "OEMs only", where either PCIe or SATA are common, standard, and amenable to individual use and upgrade...

Re:Drat

[afidel]

Even the Fusion I/O cards with SLC only push 500-700MB/s depending on the workload and they cost $7,500 for a 160GB card, SATA 6Gb should be plenty fast for a consumer standard.

Re:Drat

[fuzzyfuzzyfungus]

Worse, it's both. Mechanically identical to a mini PCIe connector; but electrically/logically identical to SATA. Won't work if plugged into a PCIe bus, because it isn't a PCIe device; but won't plug in to virtually any SATA connector; because it has the form factor of a mini PCIe card.

Re:Drat

[Rockoon]

OCZ has 740MB/s cards for an order of magnitude less (Save $7000 and spend only $650) than than Fusion I/O's offering, and with 50% more capacity too (240GB card)

For cards in the price range you are talking about, OCZ delivers 1400MB/s on its 512GB card.

You seem to be less informed than you realize.

Re:Drat

[afidel]

Are those the best case numbers or worst case? OCZ has a history of claiming huge numbers and terribly under-delivering. Oh, and at least for my use case MLC is a non-starter so the only OCZ card I'd be interest in is the Z-Drive e88 R2 which is ~$10k so 30% more for a two card solution (RAID1) and I only need ~120GB for the OLTP tables.

Re:Drat

[sr180]

Yes, we've been evaulating the OCZ Cards - and they are much slower in real life then the benchmarks suggest. Note that the FusionIO has a FusionIO Duo - which pulls 1.5GBytes a sec. This seems to be the holy grail of speed atm.

Re:Drat

[Khyber]

RAID

Good SandForce

As long as it's not on Sandy bridge, with it's gimped PCI-E, maybe you've got a point.

Re:Drat

[owlstead]

Well, that should at least allow notebook manufacturers to use the same physical design if they decide to switch to a PCIe interface. For the current generation (and probably the next SATA-3 generation as well), the SATA standard is fast enough. End users won't notice and more importantly, it won't influence the BIOS or operating system at all.

Re:Drat

[fuzzyfuzzyfungus]

As I noted there this form factor has nothing to do with intel particularly, nor did they come up with it.

I just strongly object to the use of an identical connector for two completely different, non-interoperable protocols. Were it some chintzy once-off by a bottom feeding netbook monger, trying to pinch every last nickel off production costs, it would be understandable, if distasteful; but the fact that they've gone and made a standard out of it, without adding so much as a cheap keying change to the mSATA version of the mini PCIe connector, pisses me off.

My displeasure isn't Intel specific; but aimed at the unmodified reuse of a connector intended for a completely different protocol. It's sloppy and user hostile.

Re:Drat

[bobcat7677]

Having worked with sets of comparable cards from Fusion IO and OCZ (IOXtreme and Zdrive), I can give this assessment:

Neither card met the published performance numbers. But the Fusion I/O card came closer to it's published numbers then the OCZ card in basic benchmarks making the Fusion I/O card quite a bit faster for raw throughput. Both cards were blazingly fast though pushing MBps and IOps like no tomorrow.

Real world performance suffered greatly with the Fusion I/O cards due to their software driven architecture. The CPU overhead was significant, even on a powerful multi CPU Xeon server. The OCZ cards did not have this problem.
The Price/performance ratio in real world made OCZ the winner overall. The competition was closest when excluding CPU overhead, but once you include CPU overhead the OCZ cards win hands down.
Support was highly disappointing from Fusion I/O. With OCZ you expect minimal support, but I expected something better from the "premium" Fusion I/O brand (and price point). Unfortunately, their support was no better then OCZ.
We originally evaluated the original Zdrive model which was kindof a rough implementation of the technology. If you are going to buy one now, avoid the old Zdrives...there are several problems with their design. The new R2 Zdrives have fixed these problems and are sold at basically the same price point for similar specs.

We eventually returned the Fusion I/O cards due to their ridiculous CPU penalty. We still have the OCZ cards, but have stopped using them in favor of normal SAS controllers with hot swap SSD drives. It's just not convenient to shut down a server and crack open the case just to replace a failed SSD...and SSDs do fail:) At this point, PCIe SSD cards seem better suited to high end workstation applications where it's not as big of a deal to crack open the box for maintenance.

Re:Drat

[Sardaukar86]

Are those the best case numbers or worst case? OCZ has a history of claiming huge numbers and terribly under-delivering. Oh, and at least for my use case MLC is a non-starter so the only OCZ card I'd be interest in is the Z-Drive e88 R2 which is ~$10k so 30% more for a two card solution (RAID1) and I only need ~120GB for the OLTP tables.

Here's another data point that agrees with you: I recently specc'd a machine for a client with a Core i7, 12GB RAM etc., including an OCZ RevoDrive. This machine is a massive beast, easily the most powerful I've worked with, yet it doesn't feel noticably faster than a machine equipped with a simple SSD such as one of Intel's offerings or even an OCZ Vertex2.

My experience with OCZ mirrors yours exactly; all mouth and no trousers. The reviews of the product don't help much because although the numbers look great, I find the devices don't get anywhere near the claimed performance by either manufacturer or the reviewer.

I'm fully prepared to accept that I might be doing something wrong but I've followed the usual block-alignment procedures etc. and have probably seen about a dozen of these things in the last year. As I understand it Windows 7 performs this alignment itself when conducting a fresh install.

Perfomance vs size

[BradleyUffner]

Why is it impressive that a smaller solid state drive performs as well as a standard size one? What does the size have to do with anything relating to these performance benchmarks?

Re:Perfomance vs size

[dsginter]

What does the size have to do with anything relating to these performance benchmarks?

Perhaps because of the whole decades of history related to rotating bulk storage? Without increases in spindle speed (and, thus, price), larger storage has always been faster.

Don't you remember the Quantum Bigfoot?

Get off of my lawn!

Re:Perfomance vs size

[fuzzyfuzzyfungus]

It isn't wildly impressive, since many of the larger SSDs are either smaller boards padded out with aluminum or plastic to meet 2.5inch size standards, or 2.5 inch boards taking advantage of relatively lax density requirements to save on board layers and fabrication expenses; but it is the case that most high-performing SSDs are doing somewhat RAID-esque stuff across their multiple flash chips. Thus, unless the design is severely gimped by either incompetence or cost constraints, larger device=space for more chips=more opportunity for spreading operations across multiple flash chips=higher overall apparent speed. For a very small device to hit high speeds, the maker is either doing some clever packaging, to get a competitive number of dice in that space, or implementing a nice controller that can compensate for not having substantial parallelism to play with, or using comparatively pricey flash that is high on the speed and density curves, rather than just doubling up on whatever is available at mainstream price points and taking advantage of the available board space.

Given Intel's formidable fab expertise and capital resources, it would not surprise me if two and three are at play here...

Re:Perfomance vs size

[masterwit]

Why is it impressive that a smaller solid state drive performs as well as a standard size one?

Is it the size of the ship or the motion of the ocean? (Sorry couldn't help myself.)
Otherwise, good point!

Re:Perfomance vs size

[Rockoon]

Why is it impressive that a smaller solid state drive performs as well as a standard size one? What does the size have to do with anything relating to these performance benchmarks?

The speed of SSD's is linearly correlated with the number of flash chips they contain, because the flash chips are operated in parallel (think RAID0, only its implicit in the design)

Smaller would usually mean less flash chips, so less parallelism.

Re:raid?

[Atriqus]

I'm pretty sure that's a SATA interface, but as it stands, your statement is valid for either.

Re:Windows

[jonbtn]

Perhaps you don't know that Windows (Vista confirmed, 7 should too) can map a seperate drive to a folder instead of a drive letter, if you tell it to. It is rather easy to do. You can even setup multiple paths for a single drive if you want.

Re:Windows

[fuzzyfuzzyfungus]

It goes back at least as far as XP, probably 2000 if you don't need the Fisher-Price skin...

Now, just to get back to the bigotry and one-upsmanship, any setup that forces the user to think about how best to allocate filesystem stuff between block devices, or forces them to commit to one inflexible configuration, is arguably underutilizing the capabilities of this sort of technology.

Machines are, unless the human really wants to, supposed to handle the grunt work(not to mention, keeping accurate track of file accesses, speed and latency of multiple devices, etc. properly is really beyond the capabilities of a human, at least in realtime).

What you really want is an FS arrangement that can seamlessly present you with a single logical volume, silently handling the details of what to commit to flash and what to platter, for optimal performance and responsiveness without the cost of going all Flash.

Re:Performance

[fuzzyfuzzyfungus]

If mounted upright, these guys would be just a little too tall for a 1U; but a 2U could fit several hundred... With the economies of scale enjoyed by something designed to be shoved into consumer laptops, a shelf or two of these little puppies could, with the right controller, make fibre channel stuff that costs a factor of ten or two as much wet itself...

Re:Windows

[Rockoon]

This goes back to DOS for christ sakes.

Re:Windows

[GNUALMAFUERTE]

Sorry to repost this, but I accidentally posted it as AC, and nobody is going to see it at -1.

I can't believe it either ... but there is a whole industry dedicated to dealing with windows. But it's the way our world works, sadly.

We create artificial scarcity, force people to use an inferior and limited technology, that has ridiculous drawbacks, and requires a tremendous workforce around it just to keep it functional. And we keep people using it even when there are cheaper, infinitely better, more reliable and future-proof technologies. The reason is simple: Through artificial scarcity, we keep the money flowing in a certain direction, we keep control in the same hands, and we create hugely profitable but completely pointless industries.

Think about it, we could be running 100% on clean, future-proof, secure and cheap nuclear energy. Instead, we rely on oil. The infrastructure that oil demands is huge, the drawbacks are incredible, we are polluting the environment, drilling the oceans to get some more black juice out of the earth at a huge risk.

We could also have moved all of our communications to ip-based networks, cutting down costs, and removing the need for so many different networks. We could have a single infrastructure that would provide us with high-bandwidth, low-latency internet everywhere, and put everything from phone calls to TV through that network. Instead, we are running different networks for each purpose, and within each purpose different networks for each provider. If we re-purposed all cellphone towers from all providers to give us just internet access, we could have 100% coverage everywhere in the world. Instead, we have huge overlapping (areas serviced by several providers), and huge areas with no coverage at all.

We could also be using just Free Software. It's open, transparent, reliable, cheap, and ethical. Instead, most people use windows. That means triplicating new hardware purchases, cutting 70% on hardware's lifespan, spending incredible resources in pointless activities like antivirus production/sale/deployment, and an IT structure several times bigger than required, not to mention all the lost time and profit due to preventable downtime.

But it's the way the economy works. It's the way the usual people keep getting richer, while keeping the majority of the world in line, quite and productive.

It's absolutely sad, but it's not just something that happens only in software, and it's certainly no accident.

Windows supported TRIM before anyone else

[SuperBanana]

I wonder how much that primitive joke of an "operating system" will derail the widespread adoption of these hybrid technologies.

The primitive joke of an operating system that introduced USB-flash based application acceleration (no such similar feature for any free operating system, and supported SSD TRIM commands before any other operating system? (OS X still doesn't and there are no announced plans to; Linux 2.6.32+, I believe, does only on a kernel level, but support amongst various filesystems seems inconsistent or not present; it's hard to tell. hdparm supports manually running TRIM using areas reported by the filesystem as free, but that's hardly equivalent to Windows, which "just works".)

Re:Windows supported TRIM before anyone else

[bryonak]

I know you're replying to a rather trollish parent, but still I'd like to remind you not to let facts get in the way of your biased presentation.

Assumingly you refer to ReadyBoost (which was introduced in Windows only around 2006): isn't that about the fastest way to trash your USB drive? Further assuming you are inclined to do so on a UNIX-like system, say Ubuntu:

- unmount the USB volume
- sudo mkswap /dev/sdX1
- sudo swapon -p 32767 /dev/sdX1
- increase swappiness to be on Windows levels so your disk gets aggressively cached (may need to tune the VFS caching too)

This has been available for decades, and it shows how ReadyBoost is mainly the marketing department "boosting" a simple technique.
Why noone has bothered to automate the above steps (as done by ReadyBoost)? First there is usually no need to at least on Linux-based systems (compare memory requirements), secondly having a pen drive stick out of your laptop all the time just to make it a bit faster is both cumbersome and wasteful, thirdly there are much better techniques on RAM-constrained machines.

As for TRIM... well, the 2.6.32 kernel has been released in 2009, there were two major Ubuntu releases with that kernel resp. a newer one, and 'discard' (TRIM) support takes 5-10 minutes of additional setup (I installed Ubuntu on a SSD MBP a few weeks ago). Granted, it doesn't "just work out of the box" (point for Windows!), but it works well enough.
Concerning file system support: the current standard ext4 and the future standard btrfs are discard-capable, as are number of the more obscure ones.
Others don't support it, but we have the same situation on Windows... only 50% of the commonly used file systems know TRIM (NTFS does, FAT32 doesn't). See, just a matter of presentation ;)

Re:Windows

[nabsltd]

Directory linking goes back to Windows 2000 but mapping c:\Users to it is a bit more difficult as the currently logged in users profile is always in use thus locking the folder.

There are quite a few ways to deal with this issue:

• You can schedule the mapping to take effect during a reboot (after copying the files)
• You can boot off another disk, copy the files and create the mapping. If you do this, you have to make sure to map to the drive letter that will be used when you boot from the first drive.
• Log in as the first created user, enable log in as "Administrator", log in as "Administrator", delete the first user, then set "D:\Users" as the profile directory. Every user created after that point will have their profile in the new directory, while "Administrator" will still be on C:, which is very similar to Unix where the root home dir is on /, not /home.

There are also tools from Microsoft designed to automate installs that will allow the mapping to be set at install time.

Re:Windows

[hedwards]

You don't really need tools for it. MS allows you to use the WINNT.SIF file for that purpose. It's also a convenient way to do all sorts of other adjustments that wouldn't work properly when done post install. It goes back to at least Windows 2000.

Re:Windows

[fast turtle]

Only thing is, it doesn't work that way in Windows simply because the damn / folder is part of the /user data folder. Due to that, you have to map each users /home folder on an individual basis and mapping more then a couple of folders/drives will slow boot/shutdown times considerably. Another issue is "Unlike *nix" MS didn't see fit to isolate the damn /root "/admin" folder from the /home "/user data" folder, meaning you simply can't relocate /home "/user data" to another drive.

Another issue is that the folder/drive mapping is on a user by user basis and the reason it slows shit down is the fact that Windows wants to cache the data from those mapped locations in the offline cache location. So now it copies all of that data back into "C" drive because the idiots at MS couldn't take a page from the *nix folks and use a design that's trully engineered for multiple users. I've got Vista (32/64) Win7 (32/64) and both Pro/Ultimate do not make it easy to do any of that.

Oh I'll admit that there may be some third party tools that could do it but they'll probably cost more then I can afford or the local mom and pop shop is willing to spend on IT infrastructure.

Re:Performance

[WuphonsReach]

What really makes me excited about that is the smaller chunks of RAIDed disk. Recovery by hot spare has always made me nervous due to the length of time it takes to repair a 1-2 TB hole in your redundancy.

Depends on the RAID type.

RAID 5 (and 6?) rebuild/recovery windows tend to scale linearly with the number of drives in the array. So a very large array with very large drives can takes hours/days to rebuild.

RAID 1 and RAID 10 rebuild/recovery windows scale with the size of an individual spindle, not the number of drives. So a drive failure in a RAID-10 array generally results in a very short recovery window, determined by the size/speed of a single drive within the array. Even on bigger 10-20 disk RAID-10 arrays, replacing a failed drive is only an hour or two of resync.

(I much prefer the predictability of RAID-10 rebuild times.)