Data Center Strategies

Intel's Enterprise X25-E SSD Performance

The last time I blogged about SSD performance I had a Intel MLC based SSD,  intended mostly for laptop or read intensive applications. Looking back at that blog, I reported pretty decent performance numbers with the X25-M Intel drive.

Christmas came early this year for me - I recently received several Intel X25-E enterprise SLC SSDs for evaluation. As an analyst I normally don't get the chance or have the time to get down and dirty, but this opportunity was too good to pass up. Besides, my career has been spent developing products and diving into details. Its hard to leave that legacy behind while looking at a box of SSDs just begging to be run through their paces. As, apparently, the only analyst to receive these drives, I felt obligated to take them for a ride and see if my previous enthusiasm was justified.

Lets get right to it:

I ran tests, using IOMeter on a 2.66GHz quad Core 2 Intel CPU, 45nm, 12MB L2 cache, with a 1333 front side bus, 4G memory, SATA II 300MB/s ports. The tests were run on a single SSD, both the M and E version, as well as the E version in a 4 drive RAID 5 configuration. Unfortunately, I don't have a decent RAID adapter (hint) so I used the onboard NVidia MediaShield RAID function. 

While I have more data, for simplicity I've plotted IO transaction rates for 512, 4096 and 32768 block sizes for random reads and writes. Using all random reads and writes provides significant stress on the SSD and is a good reference point for comparison to HDD performance.

Take a look at the graph:

In the graph, I plot the transactional performance of the X25-M, X25-E, X25-E in RAID 5 and a SATA HDD as a function of block size.

Its worth pointing out that the tests I ran are far from real world, but they do highlight performance under extreme conditions. Measuring performance can be a tricky business, but I believe the tests I’ve run are a good reference point and easily repeatable – except for a weirdness that I’ll point out in a few…

Take a close look at the results. The performance for the X25-E is very compelling. For random reads, the X25-E's (as a single drive and RAIDed) performance tops out around 12,000 IOPs as does the X25-M. You'll need to look closely to see the plotted lines as they overlap at the top of the graph. I suspect that the drives are capable of much more and are bottlenecked by the upstream motherboard and driver stack limitations. I didn't spend much time tuning my system so I suspect that the read number could be far higher. In any case, the values leave the poor SATA HDD in the dust.

The random write performance is equally compelling for the X25-E, operating far faster than the X25-M and making the HDD look like a stone.

The "X25-E RAID5 - Write" test, using 4 disks, stands out like a turkey in a chicken ranch. The RAID performance is actually worse than a single disk. Hmm, why is that?

When doing writes in a RAID 5 configuration, an XOR operation is required (not so when reading). Since the RAID function on my motherboard is driver based, no doubt my system is the bottleneck. This limitation does point out the stress placed on RAID adapters when dealing with high transaction rate devices. Most RAID adapters are best suited to dealing with single threaded devices (e.g. hard disks) operating at hundreds of IOPs not thousands of IOPs as SSDs can do. I'll have to wait to get my hands on a decent RAID adapter (hint number 2) before this can be explored further.

but there is some weirdness, look at the following graph:

As I prepared to collect performance data, I ran the random 4k block write test a few times. I noticed that the result varied over time and depended on the state of the SSD before the test was run. That's weird. With a hard disk, performance is very predictable and constant over time. Apparently not so for an SSD. I think we knew this but the graph proves the point. Before the test, I had conditioned the X25-E with 64K random block writes. Not scientific, but the results shown in the above graph are curious none the less. The random write performance varied four to one over the period of 30 minutes where I collected performance data at 5 minute intervals.

While much more performance testing and analysis is needed, such as the examination of latency values, I'll leave that to others with more time on their hands.... 

The performance of the Intel X25-E is remarkable compared to a hard disk. Unfortunately, the unexpected performance variability was a surprise and adds a new dimension to interpreting performance data.

Oh, and btw, the X25-E hardly got warm to the touch throughout the testing. So while I don't have a way to measure power, the X25-E clearly uses far less power than my SATA HDD that I can use as a donut warmer.

So this brings up a good point, and I'll end the blog on this note:

The industry needs a standard way to test SSDs. Period.

Please feel free to comment. 

Posted by Gene Ruth

SSDs based subsystems are getting interesting...proceed carefully

Two interesting things happened in the computer storage biz on Monday: Sun announced their new "Amber road" storage products and Violin Memory, a small startup, introduced a high performance pizza box storage appliance. Both these products are leveraging NAND flash SSD technology. Both in interesting ways.

BAM, new era...sorry couldn't resist.

It would be easy to go all gaga over these products but caution is advised. Both are innovative in their own right, both challenge the status quo of the data center class storage system business. These are innovative products. No matter what my comments to follow are, keep this in mind: these products are integrating new technologies in new ways and therefore, regardless of the technical prowess of the suppliers, caution should be exercised before inserting these products into a data center on a large scale.

Try'em to verify'em.

In my past blogs about SSDs, I've mentioned that flash SSDs are about transactional performance not capacity.  I stated that capacity will have to wait until flash memory density increases and pricing decreases dramatically - we will be waiting for awhile. So, I went on to suggest that a storage system based on SSDs, to be well rounded, must offer high performance but solve any capacity shortfall by including high capacity SATA disks. To make that workable, some magic is needed to move data around appropriately to match the data's dynamics.

That's what the new Sun storage product claims to do.

Sun, by leveraging their open source ZFS file system, with some crucial tweaks, marries SSD technology for performance and SATA hard disk technology for capacity. Compared to an all HDD system, Sun claims up to 3x read transactional performance, 5x less power, at about the same cost.

Not bad.

Turns out Sun uses SSDs tailored both for write performance to handle an internal logging function and then read optimized SSD to act essentially as a really large cache. All wrapped around their open-source ZFS file system hidden within the storage subsystem. Sun is demonstrating technology leadership by tightly integrating SSD technology into a complete storage subsystem - yes others have done a pluggable replacement for a HDD, but that's fairly obvious and less then optimal. And there have also been some demonstrations such as IBM's Quicksilver science experiment based on Fusion-io, but no significant shipping product.

There are certainly other nice features in the Sun "Amber Road" products such as an improved management interface, analytics and such but blah, blah, blah. The industry expects those things, they are the price of entry and are the nuts and bolts of any system. Of course you need good pricing, support, quality and the like - that's all good, but its expected. If a vendor does not live up to the expectation - they get voted off the island - fast.

Sun has a window of opportunity here to make some market share gains, before other major vendors show up to the party. Its on you Sun. Price the product to move, forget premium pricing. Get support right, and accounts will be won. Don't forget: others will arrive at the SSD party soon as well. Grab market share while you can. Show customers how Sun is setting a new price-performance bar only reachable with SSD technology. Game on. Watch your back, I expect strong competition in the SSD space over time, the vendors who don't react are no worry since they will not survive in the long run.

Back to Violin. Violin introduced a 2u (that's about 2 pizza boxes) flash based storage appliance. While this is not an holistic performance and capacity solution, it does offer an excellent example of what happens when you take the "D" out of SSD. No longer constrained by the HDD form factor, the Violin 1010 uses memory card like flash modules that plug into a motherboard. As part of the design, there is a RAID like redundancy for the flash modules with hot plug support in case one fails - but I gotta admit, I wouldn't touch the thing while its running. Find an intern to do it and if things go wrong make the intern the scape goat.

For those running transactional applications, like a database, this product could be a godsend. But I fear the pricing may be too high and you do need to accept working with a newbie in the space. The Violin 1010 attaches directly to PCIe (it also does FC and Ethernet) . PCIe is needed to achieve high iops - fibre channel and scsi in general add too much latency but that's the subject of a future blog.

The Violin product is interesting on its own but consider: Combine the Violin 1010 with a bunch of 1TB SATA hard disks and Sun's open storage software with ZFS. Add in a few tweaks for auto-tiering, do some soul searching about pricing and you could have an awesome product. 

So where is all this SSD technology going? Into data centers, sooner than anyone has imagined.

Fall SNW 2008 Trip Report

get ready for some rambling - lots to cover....

For those unfamiliar, Storage Networking World is a Computerworld magazine and SNIA sponsored event. Held twice a year in the US. The largest storage-only event. This is a vendor love fest, designed to bring end-users and vendors together to talk things over, get educated and generally develop a sense of the storage market for businesses. Definitely not for consumers.

As an analyst, at SNW I spend my time meeting with vendors, to hear their latest, discuss hot topics and encourage them to address customer needs. It's also a great opportunity to talk with end-users to discuss what's of interest to them.

Prior to attending SNW, I recently posted blogs on SNIA's SMI-S and SSDs and was interested to hear views on FCOE as was my Research Director Drue Reeves.

Frankly I was unprepared for the intense reaction I got regarding my SMI-S blog - I could hardly walk around without someone bending my ear about it - but more on that later.

Here's a quick drive by of what I learned:

FCOE:

This turned out to be an intense topic - I was not expecting the passion on this subject. In a meeting with Cisco, it was clear that they are all over it and pushing it hard.

We at Burton Group have been somewhat tepid on the FCOE subject. Yes, FCOE helps bridge the gap between an FC and Ethernet topology, but iSCSI offers a better price point for new installs. There is a place and value for both technologies.

Intel (they've got a 10Gig adapter) expressed an agnostic view on the subject, supporting both FCOE and iSCSI. Intel is definitely in the "let's see what develops" camp providing product support either way. Of course Emulex is all over the FC/FCOE market with their adapters. Netapp, bravely leading the market from the FCOE target side and taking a "let's see what happens view", will be delivering an FCOE target - nothing new here just reiteration of past activity.

Mostly I heard - "let's see how it goes". For the risk adverse, its still early for FCOE. Expect to see credible full implementations the second half 2009.

So we'll see how FCOE adoption goes - the market will decide. To hear more of the Burton Group view, jump on a plane to Prague and attend the Catalyst event starting this week.  Good luck with the airplane ticket prices!

SMI-s:

Based on my SMI-S blog, at times I thought I might need a security detail to escort me around the convention ;-} I had barely left the registration desk before being hit up for a discussion. My comments are meant to be constructive. There are some inconvenient  truths to be dealt with here. After meeting with the leadership of the SNIA board, it's clear that it's time for SNIA folks to huddle and develop a SMI-s game plan. A restatement of the goals for SMI-s would be a good first step. SNIA is teeming with smart people. Best wishes.

Storage Products:

Storage is getting so complicated - you'd think that it would be getting simpler. So many vendors, so many product variants. The product overlap is overwhelming.  There are lots of great storage products out there, it must make customers' heads explode...hmmm bad image...

Microsoft told about where they're taking DPM. They said that ...opps can't talk about it but take a look at this blog and look for significant enhancements. And Sun let us know about their plans for ...opps can't talk about it but look for leveraging of the Sun open storage initiative and the application of SSDs...IBM ran through their SAN volume controller for mid-range businesses - a full featured yet less performing version of their enterprise product. HDS has a new mid-range truly active-active 3Gb/s SAS disk array. Xiotech with their over-the-top marketing sung themselves praises. Hats off to Xiotech for having an actual customer in the briefing. Riverbed is making noise about getting into the storage biz with an appliance that does inline dedup, consolidation and WAN acceleration - I'm not sure how to categorize it. Bluearc has a nice story for performance NAS using TMS's SSDs as a high performance tier. And finally F5-Acopia  describe their existing network based virtualization appliance to me.

All good stuff, too much to cover in this blog.

Drum roll please..*************BAM

SSD's

My favorite subject. You may recall in my SSD related blog I said, "we need storage subsystem providers to start shipping product with SSDs in them!" Products are starting to arrive. Yes, they are somewhat brutish and much optimization lies ahead, but the game is on.

First, Intel announced availability of their enterprise SSD. Awesome performance. Small capacity. But a great start. Intel continues to legitimize the SSD enterprise offerings. Yes, kudo's to the other suppliers out there. Request to Intel: please help to standardize the spec'ing of SSDs. Performance, power, wear, bit error rates and failure rates need to be addressed. Intel, you know what to do.

The vendors offering SSD's in their products increased dramatically. Joining the existing SSD gang, EMC, Fusion-IO, TMS are vendors that either announced product or intent: Compellent, Verari, Sun, Wasabi and Rackable. More are coming - I just can't spill the beans just yet. If I missed someone, please comment on the blog - the list grows every week. Lots of rumors out there. IBM's showing a SVC-based million IOP beast using Fusion-IO under the covers, expect this to be productized as well.

Of course, HP now ships blade servers, not to mention laptops, with SSDs. How long will it be before SSD chip sets end up on a server motherboard?

For Burton Group clients, watch for my upcoming in depth research document on SSD's.

I declare the SSD games open!

But I would call these Novelty products - designed for extreme IOPs, not so much a blended IOP-capacity capability and, unfortunately, with premium pricing. Close but not yet where the market needs to go.

Here's what the market will really love: a blended system with SSDs for performance and terabyte SATA disks for capacity. To make this work, auto-tiering will be needed under the covers, transparent to users. Ideally, this product would allow policy-based data movement leveraging usage patterns and storage costs. Compellent is uniquely positioned to do this, but I fear their infrastructure is not yet optimized for SSDs. SUN's got some interesting ideas using ZFS. And, Wasabi, a small but innovative vendor, is combining SSDs with an object-based file system, which theoretically can easily identify and move data objects across performance tiers.

And to drive down SSD subsystem costs, the "D" must be removed from SSD and replaced with "PM" for persistent memory. Repackaging SSPM's within a storage subsystem will greatly reduce costs and allow for performance optimization, not unlike what Fusion-IO and Violin already conceive.

Exciting times. But be patient - this may take years to unfold...development cycles can be long...and, of course, whatever we end up with will likely be complex beyond comprehension.

Perhaps the geniuses who dreamed up credit default swaps can help out ;-}

Let me know what you thought of SNW by commenting on this blog. Thanks!

Posted by Gene Ruth

Has SNIA’s SMI-S failed?

“We’re not enamored of truth. It is too often painful, discouraging and it tends to undermine our self-image. We prefer comfort. Reassurance. Well being. Good Cheer. Naked optimism. Nobody wants to hear the facts when they clash with a happily imagined reality. It is after all, a terrible thing to be the only person in town who can see what’s really happening" [from “Odyssey” by Jack mcDevitt].

To IT data center storage vendors: Are you fooling yourselves about SMI-S? Is SMI-S relevant anymore? Is anybody really implementing it? And where is that mystical standards-based storage management console? All difficult questions and, I fear, with unpleasant answers. It’s been over 5 years since SMI-S was first kicked off. The truth must be faced. I’m not seeing truly complete standards-based storage management implementations. I mean implementations that completely manage a storage system or device within the confines of SMI-S, even assuming a few vendor unique extensions – which, btw, are allowed by the spec.

At this point, you may be scratching your head and wondering what SMI-S stands for. If you don’t know, stop reading this blog unless you relish the anticipation of a possible train wreck. For those who do, you may be wondering about my sanity or nodding your head in agreement.

Let us digress. For some education, go to this link and take a look at the SMI-S document. I'll wait. It might be awhile - the spec is over 1000 pages.

SMI-S is very complicated with a lot of loose ends but it CAN BE MADE to work.

The promise of SMI-S is simple: By getting storage vendors to instrument their gear in a standard way, theoretically, a single management infrastructure with a single console could manage multi-vendor storage environments in a common way. Sounds great doesn’t it? We all dream about this interoperability.

Trouble is, no-one-bothers in a serious fashion.

Sure, some of the top tier players - two, three, and four letter names – do implement some of SMI-S in their products. But it’s the “some” part that’s the rub. I frequently tele-conference with enterprise class storage subsystem vendors– large and small – to hear their latest, and I usually ask this question: Do you implement or plan to implement SMI-S? If so, how much and when? …Far and away, there’s a pause, I hear some shuffling and throats clearing, and get one of two answers: “we are watching that space and will provide an implementation when our customers ask for it”. In other words: forget about it.

Or I get told: “Yes, we do!”

And I ask: “Completely? Does that mean I don’t need your management application to configure and monitor your gear?”

And they respond: “Well, ah, no…”, again paper shuffling and whispering….

Back in June, Burton Group held its annual Catalyst Conference. During the conference, Drue Reeves (VP & Research Director) challenged the industry with his “count-down to interoperability”. He asked, “when will truly interoperable, standards-based storage management applications and instrumented systems be available?” He gave a one-year time frame – I fear he will be disappointed.

So has SMI-S failed? Will vendors seriously implement it? Vendors please let me know. Tell me you are nearly there. After all, as a matter of confession, I and others at Burton Group contributed to the spec and know many very smart people who have worked long and hard on it.

I want to see it succeed.

But what do I tell my clients asking for a unified storage management environment?

If SMI-S has failed then what went wrong? Did vendors realize or anticipate that cooperating to make their equipment interoperable - at a management level - hurt revenues? Did storage systems get so complex and feature rich, that the modeling approach taken is too detailed? Or, is it simply that SMI-S is too complicated and vendors don’t see the payback for investing precious development dollars on it?

I suspect it’s all of the above.

Perhaps a new approach is needed.

Enterprise class storage subsystems have become feature rich over the past few years – what top tier vendor doesn’t do thin provisioning, tiering, replication, etc. All these features are uniquely implemented and a cause for vendor differentiation. The real value of these new features is the delivery of a Quality of Service (QOS) such as data protection, scalability in the face of data growth, performance, and the like. Unfortunately, SMI-S focuses on the down and dirty modeling of the internal specifics of these features – the place where vendors innovate and add their secret sauces.

I’d suggest a new approach: model storage management based on QOS or, more simply, service delivery. For example, rather than modeling the intricacies of thin provisioning or RAID sets, instead, through standard api’s, identify that the equipment supports specific features characterized by key standardized parameters associated with each feature. Why should users specify what disks go into a RAID set? Instead, let them define the level of redundancy and expected performance, letting the equipment figure it out from there. Some vendors already do this in a proprietary way – it's not a new idea. This can be conveniently modeled with XML.

So, back to the main question: Has SMI-S failed? What do you think? Is there hope or should the storage industry rethink how to achieve interoperability in IT data centers or, more to the point, do vendors want to do it to begin with? Where does the SNIA board stand on this issue? After all, the members represent the vendors whom this blog is addressed to.

I know one truth for sure: IT data center customers want storage management interoperability and they want it now.

Posted by Gene Ruth

SSD performance numbers

There’s a lot of hype surrounding Solid-State Disks right now, and if you’ve read tech news recently or my last post, you’ve probably stumbled upon some of the impressive numbers coming from the SSD camp. All in all, it looks like SSDs are faster than Hard Disk Drives(HDD). Great…but how does that affect you? Does this mean you have to replace your HDD? What about the servers for your business, and storage, and, and, and…? Well, obviously, I’ve been asking these questions too, and (lucky for me!) I have a X25-M SSD from Intel to evaluate. Let’s see if it holds up to the hype.

First, though, let’s put things into perspective. Much of the SSD hype stems from the fact that over the past 10 years CPU performance has increased 30 fold; whereas, disk transaction performance has been progressing much more conservatively and deliberately. Look at the SSD innovation from the perspective of the theory of evolution. Way back, organisms swam in the ocean, made progress through the years by adjusting fin sizes and shapes, but didn’t see real change until that one fish decided to turn those fins into legs. Then Bam. New era.

Or, take the Humanoids running around an ancient continent. They made slow and deliberate progress using that opposable thumb to hit each other with sticks (remember the opening scene from 2001: A Space Odyssey?), and then one day there’s an evolutionary leap, and we have fire. Bam. New era.

According to the hype, disk technology is making a leap to a new era. Bam. Solid-State Drives. Bam, IBM announces 1 million IOs per second using Fusion-IOs flash based storage.

Our present hard disks have these rotating parts, which, let’s face it, are so Industrial Revolution. The new SSDs with their new FLASH technology seem cool. But just because something seems good doesn’t mean it’ll last. And just as with species evolution, it’s survival of the fittest. Even some promising candidates meet their end. I mean, what ever happened to the giant, prehistoric Megalodon (it was pretty neat)? The Neanderthals? Or the lead actor from 2001: A Space Odyssey? (Okay, maybe he’s still around, but you get the point.)

So I’ve taken the X25-M SSD that – mind you – is intended for laptops, and gave it a try running Windows Server 2008 with hyper-V to see if it really will improve our lives – or better yet, YOUR business.

Here’s the setup – quad core Intel proc with 1333MHz front side bus and 800MHs DDR2 memory. Now the SSD is only 80GB and I wanted everything to run from it. Not exactly a configuration that I would recommend but it’s the best I can do until I get another SSD (hint…). I loaded the Host OS onto a dedicated 25GB partition then, created another partition for the VMDKs. I created two virtual machines – an instance of Vista and another one of Windows Server 2008. Trying to keep it simple I did not load anything else. So the question I asked myself is – will I notice a difference or will I need to tease the difference out. Short answer – no teasing necessary. Take a look at the chart.

The time to boot a vanilla Windows 2008 server went from 53 seconds to 22 seconds – a greater than 2x improvement – not bad. I achieved similar improvements for the virtual machines running on top of hyper-V. For the host boot time, I measured from when the BIOS was done booting until I saw the login screen. For the virtual machines, I measured from the time I clicked the start button in the hyper-V manager until I saw the login screen.

Emboldened, for you techy types, I ran some down and dirty transactions measurements using IOmeter. In the following graph note the huge differential between the HDD (a Seagate Barracuda SATA 500GB) and the Intel SSD. I ran random IO, with 512, 4k and 32k block sizes. The SSD, worst case, has three times the write transaction rate of the HDD. Best case I measured over sixty times the performance. Again, no teasing necessary. For detailed measurements look at this site and this site.

Now imagine numbers like this in your data center. You’re at work, cup of coffee in hand, looking at your newly installed SSDs, and not only are you marveling at the increased performance, but you notice you can even hear the birds chirping outside, rather than that insistent hum from the old HDDs. (Okay, maybe a bit much, but you get the point.)

According to the numbers, do I think SSDs will help improve our lives? Does it make sense to put SSDs into servers? Yes and Ye--…well, maybe. With the present price structure, the advantage to the SSDs is entirely dependent on the application. For example, data centers that rely on high performance will find the SSDs as a must have. On a base level, SSDs are able to boot more than twice as fast as HDDs by my measure, and when dealing with a large number of servers, this makes a substantial difference. However, for companies that rely on greater amounts of storage for movies, music or other types of large files, SSD might not be the way to go…at the moment. On a transactional performance basis, 10 SSDs can replace hundreds of HDDs, but on a capacity basis the cost per gigabyte does not make the switch cost effective with the present price structure. For SSDs to be used with applications requiring gobs of capacity, we’ll have to keep rubbing sticks together hoping for fire.

Neither SSDs nor HDDs seem to be going the way of the Megalodon anytime soon. For now, the incorporation of both devices might be the optimum solution for business needs as the SSDs can increase performance in certain areas all the while balanced with HDDs for cheaper storage per dollar.

Oh, and one not so small thing: we need storage subsystem providers to start shipping product with SSDs in them!

Overall, the X25-M SSD seems to hold up to the hype. But before giving my final conclusion, I’ll have to report on how quickly World of Warcraft loads…that is, unless I get caught up playing it.

Posted by Gene Ruth

SSD - Storage is just the first wave

Solid state disk (SSD) made from FLASH memory technologies are suddenly all the rage, with big vendors like EMC and SUN giving details on their strategy for using SSD to improve I/O performance in enterprise storage. It's not surprising that storage is the first to benefit from the technology for two reasons:

1. Performance: Conventional disk cannot address the growing gap between storage and overall system performance. You can't just make disks spin faster or the read/write heads move more quickly, physics gets in the way.
2. Ease of use: FLASH-based memory packaged as disk (i.e. using the same 3.5" and 2.5" form factors and interfaces) is pretty easy to integrate with existing storage systems, it doesn't take a rocket science degree.

The immediate impact will be on sales of high performance SAS and FC disk, where performance is prized over capacity. Why buy (and subsequently power and cool) ten 72GB 15K RPM FC drives if one or two SSD drives can give you better performance and still meet your capacity requirements? Yes, I know the technology isn't quite there yet, particularly for write performance, but it will be, and it will decimate the high-performance disk market.

With SSD in disk drive form factors a foregone conclusion (my bet is most enterprises will have it for at least some applications by 2010), the more interesting question is how the technology evolves and it's long term impact on several areas:

• Server form factors: Some of the form factor of a typical server or blade is governed by the number of drives supported. But why have drives at all, FLASH memory doesn't have to be packaged that way, it could installed in slots on a motherboard for example. So FLASH bring about new more dense server form factors as storage performance, power consumption, and storage form factors change.
• File systems: All the file systems in widespread use today are designed to overcome or at least mitigate the inherent limitations of conventional disk. For example many file system optimize placement of disk blocks for a particular file to minimize disk head movements caused by fragmentation, others employ defragmentation tools to address the problem. But FLASH-based storage is truly random access, performance is the same for every single block on the device, rendering the whole concept of fragmentation meaningless.
• I/O interface protocols: Protocols like FC and SAS are designed for use with conventional disk, but they don't make sense for high-performance I/O from memory, which would be better suited to a pure memory type interface without all the storage protocol overhead.
• Main memory: At first glance this one doesn't seem that attractive, after all FLASH is much slower than DDR2 DRAM isn't it? But it doesn't have to be that much slower, as Spansion are demonstrating with their NOR FLASH-based DIMMs that plug right into the memory slots on the motherboard. Read performance is not quite up to DDR2 speeds, but it's close enough, and 32GB/DIMM (that's 256GB in eight slots) is very attractive for applications that primarily read from a database that can now be loaded into memory.
• Operating systems: Changing the performance ratio between storage and compute may open up new possibilities in OS design, for example the paging system could be implemented on FLASH storage interfaced directly to the memory bus.

I could go on (and probably will do in the future :-), but that will do for now.

Posted by: Nik Simpson

The Second Coming of the Hard Drive?

Recently, there has been coverage in the IT press of an announcement of work being done in Holland on a technology called Heat Assisted Magnetic Recording  (HAMR) for disk drives. In short, HAMR involves using a laser to write data on a conventional magnetic disk. Almost without exception, the coverage has focused on a claim that a lab experiment was able to change the magnetic state (i.e. whether it's "1" or "0") of a bit on a magnetic recording media, 100 times faster than conventional hard disk technology works with headlines trumpeting "hard disks get 100 times faster" which sounds great. Of course it's not true!

First, this a lab experiment working with magnetic media with storage densities roughly 1/10th of what we currently have and is 10-15 years away from a production device, so don't hold your breath. More importantly, a hundred fold improvement in write speed for a disk will have no perceptible impact on hard disk performance, for the simple reason that drives today are not limited by the speed with which bits can flipped. Today, a hard drive is capable of around 200 discrete I/O operations a second, that means on average any I/O takes around 5 milliseconds to complete. Nearly all of that time is used moving the disk read/write head to the appropriate location on the hard disk. To see how just insignificant the proposed performance improvement is, lets look at the claim:

The new technology can flip a bit in 4x10^-14 seconds, today, the figure is more like 4x10^-12 seconds

That almost sounds impressive until you look at it in the context of the time to complete the I/O operation of around 5x10^-3 seconds. We are talking about shaving a few billionths of second off an operation that takes several thousandths of a second to complete. So, no need to get the party hats out just yet! At the same time, if perfected the technology would lead 2.5" drives that can hold 40-50 TB of data, about 100x times more dense than today's drives. At the end of the day, only extraordinary improvements in drive mechanics to reduce seek time and rotational latency will have any detectable impact on disk performance. The problem is that, for drive mechanics we run into the dictates of hard physics such as inertia (speeding up head movement) and centripetal forces (spinning the disk faster) that can't be worked around easily.

No amount of improvements areas of drive performance that don't reduce seek & latency penalties will have any significant impact on real world I/O performance. The reality is that hard drives are Dinosaur technology that has to be replaced before we'll make any significant headway in addressing the storage performance gap, and so far the progress on technologies such as holographic storage have been glacial, while flash-based solid state disks making their way into laptops are a long way from being practical in enterprise storage applications because of the longevity, cost, and storage density of the devices.

Meanwhile, if you are looking to increase I/O performance today, the only proven ways are:

• Cache I/O operations in memory (both in the host and the controller)
• Solid state disks for highly I/O intensive applications
• Stripe a logical disk over many physical disks to get parallel read/writes

posted by: Nik Simpson