Extensible NetApp: Thinking outside of the box

Shed a tier with PAM II: The SSD tier

The DAS disruption series has been driven by a belief that expanding memory sizes change the fundamental behavior of storage systems. What I’ve argued is that expanding memory sizes put an increasing premium on the ability to do efficient write operations and an increasingly low premium on the ability to do read operations. And I have argued that the future belongs to the Better Than Fibre Channel Array vendors who have optimized their systems for write performance.

It’s always gratifying when your company releases technology that buttresses your point :-)

The tier to shed isn’t FC, it’s SSD!

And just to be clear, the point of this post isn’t to argue FC is dead. because that’s too easy. The point of this post is to argue that Flash as a stable storage medium is dead.

The PAM II is NetApp’s Flash memory card. Unlike SSD’s that can only accelerate data that is stored on the SSD, the PAM II allows NetApp to cost effectively extend the buffer cache for all data in a system. Because the PAM II is a buffer no expensive complex data distribution system like FAST is required.

The PAM II allows a customer to add 256 512 GB of Flash memory per card to the system buffer cache, reducing the total IOPS that the disk subsystem has to perform across the entire dataset stored in the system. The PAM II and the surrounding software architecture allows NetApp to leverage flash for hot data, while leaving disk to store cold automatically.

There are lots of ways to make the point, but this particular set of graphs makes it quite eloquently.

Using hardware that has worse latency than FC, the latency of the storage system was about the same with fewer spindles for all of the data.

What’s going on is that the although the total number of write operations has stayed the same, the total number of read operations from disk have gone down. Because the read operations are being served from high speed flash, the total latency actually delivered from the storage device is better than a system that uses FC drives.

For those  who followed our previous announcements on the PAMvI, this is no surprise.

And now for more mission critical workloads…

What should be more interesting to folks is that we’re able to achieve similar results for OLTP workloads (aka mission critical).

In a recently produced white paper Using the Performance Accelerator Module II in Online Transaction Processing, NetApp demonstrates that a PAMvII is equivalent to doubling the number of disks:

 

And at the same time improving the system response time:

I’ve said it before, and I’ll say it again, Flash is a part of the volatile memory hierarchy, not the storage hierarchy. Flash is most efficiently and effectively used as a cache not a solid state device.

Technorati Tags: emc,netapp,flash,ssd,emc fast,storage performance,tier 0,tier 1,pamvii

The DAS Disruption part 11: The end of big iron storage

For a summary of the story so far, I suggest you start here.

Regular readers of this blog have to endure the periodic “EMC sucks, NetApp rocks” post. Today, I want to applaud EMC for what was a remarkable transformation from selling exclusively big iron storage to selling a mix of big iron and modular storage. That transformation, however, sowed the seeds of the destruction of highly available tightly coupled storage infrastructures.

In the 1990’s, EMC’s Symmetrix was the benchmark for available storage. EMC, the company, provided exceptional service, and the hardware did the same.

Like the server big iron to which it was connected, the Symm allowed application and OS developers to assume that every disk block had the same degree of worst case latency, even when hardware started to fail.

The Symm was able to do that because it was engineered to hold a certain amount of capacity and the infrastructure necessary for the performance and availability was baked into the hardware and cost structure.

In 2002, however, well into the disruption of server big iron with x86 servers,  EMC hit a brick wall. Customers weren’t buying monolithic highly available servers, and they didn’t understand why their cheap servers needed to be connected to big storage iron.

One of the beneficiaries of that disruption was NetApp. We convinced many customers to replace their x86 windows file servers or Solaris file servers connected to Symm’s with our FAS systems. 

Faced with what was essentially a customer revolt, EMC responded with the CLARiiON. The CLARiiON unlike the Symm has a modular storage system design that has less availability and fewer performance guarantees than a Symm. But it is cheaper.

It’s cheaper because there is a tradeoff between uniform resource latency and availability and cost and modularity.

The transition, watched from the outside, is nothing short of breathtaking. Within a few short years, the big iron company became the modular company while still maintaining it’s lead in big iron.

The seeds of destruction

The problem with that transition from big iron to modular storage is that it exposed to the world the notion that storage wasn’t a uniform resource.

With the Symm, application availability was tied to the Symm. And since the Symm provided uniform performance, no one cared what was going underneath the Symm. With modular storage, we introduced all sorts of preferred, and optimized and unoptimized notions of data access, making it obvious that storage wasn’t quite so monolithic ..

Well once you tell people that it’s no longer cost effective for storage to solve the whole problem yourself, folks have to ask: how much of the problem do they really need storage to solve?

Put differently, modular storage systems provide high availability but not uniform access to the underlying storage.  The question then becomes, how non-uniform can the access be? In other words, how coupled does the storage have to be? And how highly available does the storage have to be for the services layered on top of it to be available?

Technorati Tags: storage,emc,netapp

Minor formatting error.

Unified Storage: The Professional Movie

After my award winning amateur performance, NetApp's marketing team decided I need to get my story done professionally.

I like to think of my first video as my screening test, this is the real thing.

In my last post there was a sense of confusion about data protection.

I said NetApp had a single mechanism for doing data protection and touted the value of that.

Some folks wondered: isn't flexiblity a good thing? And I agree. What I meant, was that  for DR, there was a single mechanism for doing data protection regardless of data access protocol (CIFS, NFS, FC, iSCSI).

If you want to hear more about our data protection, I recommend this video

Versatility of SnapMirror -- NetApp -- Storage Efficiency

A video of unified storage

After blogging about the value of unified storage, I thought a video would be nice…

FAS the new DAS

When the Forrester report came out with its report on DAS, the response from the storage industry was predictable:

1. Same old, same old
2. Immature, unreliable, inefficient technology
3. Whole product has value

And when I read those responses, I could not but feel a little bit like every disruptee feels when a new disruptive technology shows up.

The funny thing with the Forrester guys is, that they are right. DAS is disruptive to the traditional storage array, in a way that the storage industry has not owned up to. And that unless the storage industry responds, DAS may win out.

The problem with the report is that they knew there was fish in one of the ponds, they just went to the wrong pond, and their report was dismissed.

So instead, I am going to offer a different point of view about the disruption of storage by DAS.

John Fulbright, has an excellent analysis where he shows how the DAS disruption is tied to RAID-5 SAN and how NetApp RAID-DP is different.

This disruption, because it is a storage disruption will take years to shake out. Storage infrastructures change at glacial rates. Once a trend, however, sticks it sticks for a long long time.

This is going to be a long series, but the short version is below.

The disruption in a nutshell

In the late 90’s and early part of this decade, Data Center Architects, reduced costs in their data center by moving away from expensive reliable server architectures to cheap and unreliable server architectures. To address application availability, the Data Center Architects moved to applications that achieved high availability by leveraging clustering for CPU and Memory availability and highly available storage for data availability.

The new DAS disruption is about providing data availability by moving availability out of the storage infrastructure. The new DAS disruption is made possible because traditional legacy arrays are wasting vast amounts of capacity because they were built to optimize the performance of servers that were disrupted last decade. The waste is so significant that the cost advantage of DAS outweighs the value of improved utilization through centralization.

What does this mean for NetApp?

Ultimately shared storage must always compete with DAS. If you can’t deliver superior value at a reasonable cost, then you don’t belong in this industry. Arguing for total customer value is a quick way to being a disrupted technology.

NetApp storage systems were optimized, interestingly enough, for the server infrastructures that are being built today. Which means, that uniquely in the storage industry, we are able to deliver performance and efficient use of capacity.

If you compare the performance, capacity and utilization of our low-end storage systems against a DAS system, it turns out that we are cost competitive and offer superior utilization, performance and availability. This is exactly what John shows in his blog.

More prosaically it means that you can offer external storage in a DAS configuration and deliver value, a surprising conclusion.

And if you can deliver value for DAS, then you can deliver the value of centralized storage if you can combine that DAS into shared storage. And that’s where our ONTAP 8 product architecture fits in.

So unlike the rest of the storage industry, my reaction to DAS is: we can do that too, and we can do that better.

FAS Deduplication is here for the long haul

The current tussle between NetApp and EMC has, predictably, created a dust-up amongst the corporate blog-o-sphere. After all the storage industry believes in death-match-competitive-blogging, none of this polite, civilized discourse, here we reach for the knives and hack!

And it would be tempting to point out that Chuck and Storagezilla are like Aesop’s fox, but that would be too easy.

However, there is a point of confusion in the storage and analyst community about the relationship between FAS dedup and inline dedup technologies like the one provided by Data Domain. There is some opinions that say that this proposed acquisition is driven by flaws in our technology. There are some customers who may be wondering whether this reflects some flaws in our technology.

So let me state again, what I said earlier:

The rest of the storage industry has adopted inline because they lack NetApp's storage virtualization layer that enables primary deduplication.

But because that confusion is real, I wanted to remind folks about why FAS dedup is unique. And more importantly, why NetApp FAS dedup is different and must be different to meet the market requirements our storage plays in than other inline dedup solutions.

I went over this in a post several months ago in a post titled A little digression on Inline Dedup.

The basic points I made are the following:

1. Deduplication is a compute intensive operation
2. Single core performance is stagnating
3. Latency isn’t something it’s the only thing
4. Backup workloads are latency insensitive but throughput sensitive
5. Primary storage workloads are latency sensitive

And I observed that because of 1, 2 and 5 any inline process is problematic if not deeply flawed.

And then in another post I observed that the important question for dedup isn’t how efficient the dedup process is, but how well do you perform after the dedup process is complete.

Our deduplication process performs well because of it is integrated into our core WAFL data structures and because, unlike everyone else in the storage industry, we’ve solved the random IO performance problem introduced by on-disk layout fragmentation.

FAS dedup is the most widely deployed dedup solution, the only workable primary storage dedup solution, realizes tremendous cost savings, and continues to provide compelling value to our customers.

Additional Information and Where to Find It

As mentioned above, in connection with the proposed acquisition of Data Domain, on June 4, 2009, NetApp filed with the SEC a Registration Statement on Form S-4 (Commission File Number 333-159722) containing a Proxy Statement/Prospectus for Data Domain’s stockholders, and NetApp and Data Domain each plan to file with the SEC other documents regarding the proposed transaction. The definitive Proxy Statement/Prospectus will be mailed to the stockholders of Data Domain. BEFORE MAKING ANY INVESTMENT OR VOTING DECISION, INVESTORS AND SECURITY HOLDERS OF DATA DOMAIN ARE URGED TO READ THE PROXY STATEMENT/PROSPECTUS AND OTHER DOCUMENTS FILED WITH THE SEC CAREFULLY IN THEIR ENTIRETY BECAUSE SUCH DOCUMENTS CONTAIN IMPORTANT INFORMATION ABOUT THE PROPOSED TRANSACTION.

Investors and security holders will be able to obtain free copies of the Registration Statement, Proxy Statement/Prospectus and other documents filed with the SEC by NetApp and Data Domain through the web site maintained by the SEC at www.sec.gov and by contacting NetApp Investor Relations at (408) 822-7098 or Data Domain Investor Relations at (408) 980-4909. In addition, investors and security holders will be able to obtain free copies of the documents filed with the SEC on NetApp’s website at www.netapp.com and on Data Domain’s website at www.datadomain.com.

Participants in the Acquisition of Data Domain

NetApp, Data Domain and their respective directors, executive officers and certain other members of management and employees may be deemed to be participants in the solicitation of proxies in respect of the proposed transaction. Information regarding these persons is set forth in the Proxy Statement/Prospectus described above. Additional information regarding NetApp’s executive officers and directors is included in NetApp’s definitive proxy statement on Schedule 14A, as filed with the SEC on July 14, 2008, and additional information regarding Data Domain’s executive officers and directors is included in Data Domain’s Annual Report on Form 10-K and Form 10-K/A, as filed with the SEC on March 13, 2009 and April 30, 2009, respectively. You can obtain free copies of these documents from NetApp or Data Domain using the contact information above.

How Flash killed the TLA

I come not to praise the traditional legacy storage array (TLA), but to bury it. The architectural model that served it so well for so long, has reached the end of it’s useful lifespan. Killed ironically by the technology that it thought would prolong it, Flash.

Sadly for many customers, the end of the TLA will be more like  a long slow death spiral, rather than an abrupt demise. Sad because those customers will spend money and time trying to make the wrong architecture work.

Why?

The traditional storage array is optimized for read operations. The array minimizes the amount of effort it has to do on a read by putting the disk block exactly where the client thinks it is. This means on a read, the array has to do a trivial lookup to get the disk block.

However, the downside is that the array makes the write operations more expensive.

When you factor in RAID and the impact of RAID on write performance, the TLA offers a poor value proposition.

But the poor write performance was okay as long as the read operations dominated.

Enter bigger memories

The problem with the TLA architecture was that it was designed in a pre-64 bit era, and in a pre-flash era. In that era, the servers connected to the storage array were memory starved. Because they were memory starved, the buffer cache on the server was quite small, resulting in more IO operations for reads.

In the new 64 bit era, and especially with the availability of Flash, the servers are no longer memory starved. You can now have more volatile memory than you ever had before. What that means is that the average server is rarely doing a read operation, but instead is mostly doing write operations.

The impact on the TLA is that the workload shifts from being dominated by read operations to a bigger mix of write operations.

The problem, then, is that the poorly optimized path is now a bigger piece of the overall workload mix.

Or more to the point, that poor write behavior  around RAID-6, suddenly becomes a very big issue.

Which is why, after all, the TLA vendor is recommending RAID-10.

Their poor write performance is forcing them to throw more hardware at the problem.

And their poor write performance is forcing application administrators to look at alternative storage architectures.

Enter write optimized storage arrays

Write optimized storage arrays, like ONTAP, are designed for this new world order. The downside, of course, is that the read operations are more expensive, but if the mix shifts between read and write, then that’s a reasonable tradeoff.

And as I’ve said before you can solve the read performance problem …

And why do I pick this date?

Because it has been almost a year and a half since the time the Better Than Real Fibre Channel guys demonstrated that their architecture outperformed the TLA.

An architecture optimized for read was beaten by an  architecture optimized for writes, and the response continues to be silence.

So goodbye TLA, you are so last decade…

10 years at NetApp!

Between writing the requiem of the 15K RPM drive and dealing with a broken sewage pipe at home, I forgot that June 1st, was my 10 year anniversary at NetApp.

The coolest part of my time here has been the people. Listing them all here would be problematic. After all I don't want to encourage recruiters.

The most surprising part of my tenure here has been two things:

1. Everybody keeps saying we're going to go out of business because we're a niche NAS player, and we don't.
2. EMC is still our most ruthless competitor. I mean c'mon guys, just keel over and die already :-)

I've said this before, but the executive team at NetApp continues to be the single most important reason I've stayed here over the years. When I joined in 1999, I wasn't sold on NetApp for the long haul. And after the thespian troupe (acting CEO and acting CFO and acting COO) that mismanaged SGI, I wasn't sold on executives adding any value. But after a day long training session on NetApp where Tom Mendoza and Dan Warmenhoven spoke, I was sold on them, and the company. And nearly 10 years later, the team they've built and continue to build is probably the best retention policy they could have.

Some random fun facts:

1. I spent 9 years working on anything but Data ONTAP.
2. When I joined NetApp I thought SCSI was a cable not a protocol
3. There were fewer than 500 employees when I joined.
4. I remember when we thought it was Network Appliance (s) not Appliance
5. I've been through bubble 1.0, recession 1.0 and now bubble 2.0 followed by recession 2.0... and I am praying for bubble 3.0
6. When I joined 90.8 million in a quarter seemed like a lot of money.
7. Dave's book is still the best and most accurate history of NetApp
8. I managed to serve in the Greek military and debug code at the same time.
9. It took me only 9 years to figure out what Tom Mendoza meant when he said that attitude was everything.
10. I've changed jobs in NetApp 3 times going from Streaming Media, to Storage Management and now data protection, demonstrating that I lack any ability to focus.
11. I've had way too Dr. Evil moments: well let's say we have 100 MBs (crowd laughs), okay 100 GB (crowd laughs), fine folks, throw me a bone, how about 100 PB (crowd stops laughing)
12. I've been here long enough to see disk replace tape (oops not yet)
13. There was a point in time when an outrageous opinion was that Flash would replace disk. Which is especially funny, because now it's outrageous to argue that Flash won't replace disk.

After 10 years, it's tempting to want to jump off the train, but man, if I can forget that this was my tenth anniversary, who knows, maybe there's another ten in me...

15K RPM Drives, Not dead yet?

EMC, to their credit, were the first storage vendor to put an SSD into their array.

One point of view, was that the emergence of SSD was the end of the high performance disk drive. After all SSD’s offer vastly improved performance over 15K RPM drives, and on IOPS/$ basis are cheaper along every meaningful dimension.

Except, I have this nagging doubt.

SSDs are a great technology. They are fast, they are cheap and they look like disk drives. The problem with an SSD is that data at rest is vastly more expensive if it’s on an SSD and not on spinning rust.

So?

As I have said earlier, the reason disk drives have been so successful in the market place is that application IOPS data density maps very well to IOPS density of disk drives.

Most of the data an application has is cold data. So paying for high performance for cold data is a very expensive proposition. If the array can provide high performance to hot data even if the data is stored on slow storage, then the value of Flash as a medium for storing the data is lessened.

In other words, yes 15K RPM drives don’t provide uniformly excellent performance, but they may provide good enough performance for enough of the data to make SSD’s a niche.

Now, I’m not arguing that SSD’s are not the future, nor am I arguing that 15K RPM drives are the past, I am just observing that the future is murky.

The assault on RAID

Earlier this week I wrote about how the core architecture of Traditional Legacy Arrays limits the performance of RAID-6. Just to recap, RAID-6 requires an additional read and an additional write over and above RAID-5, limiting it’s performance and applicability.

What I didn’t do is explore the consequences of such limitations.

The first, and most obvious consequence is the proliferation of RAID-10. SATA disks are too big to be used with RAID-5. The only alternative is to use RAID-10 when you need both resiliency and performance!

This singular fact turns the economics of RAID on it’s head.

Let’s be clear, the reason RAID became so popular is because it solved two generic problems: performance and availability at a lower cost than the alternatives. The problem with disk drives is that they have a limited number of IOPS per spindle. The only way to get more IOPS is to combine more of them and spread the IOPS over the spindles. The downside is that the MTBF (mean-time-between-failures) of the disk drives makes the group of disk drives increasingly less reliable.

So if you’re a customer that has business critical application that requires availability and performance, you’re more or less screwed. To get the availability and the performance you need to 2-3x the storage. As an example, consider GFS, that uses mirroring and no RAID. GFS will have at will have at least 3 full copies of the data.

Enter RAID.

RAID is so common place now-a-days that we forget how miraculous it must have appeared to folks when it first showed up. RAID allowed the computer industry to provide resiliency and performance at a lower cost! So for example, using RAID-5, you only need 1/5 (4+1) additional storage to get terrific data resiliency compared to the 3x that GFS requires.

Enter Big Drives

The problem with RAID-5 and RAID-4 is that the probability of various kinds of failures increases as the disk drive capacity increases. No surprise.

NetApp pioneered the use of RAID-6 with our FAS systems, and because of WAFL we are able to preserve the RAID value proposition of increasing capacity and performance with modest increase in physical capacity.

But we’re not the market.

As much as I would love to be the top-dog in the storage industry, the Traditional Legacy Array vendors, and in particular EMC, don’t have a very good RAID-6 implementation. Or rather, as I said earlier, their core architecture precludes a RAID-6 implementation that they can recommend for all workloads.

And because of their share of the market, they define the technology expectations for everyone who intends to use storage.

The problem is that we don’t live in a static universe. Software companies, like Microsoft, are continuously trying to reduce the total cost of their solutions. And although it’s nice to be able to bill customers for RAID-10, the reality is that for every dollar that goes into the hardware, Microsoft makes less money.

So Microsoft and the rest of the software industry, faced with the reality that the dominant Traditional Legacy Array vendors, were pushing RAID-10, or basically 2x the physical capacity for each byte of data stored, looked elsewhere for a solution.

It must have been particularly galling to software vendors that disk drive capacities were increasing, disk drive cost per GB per decreasing, but the storage infrastructure was increasing in cost reducing the total available pie.

Enter host based replication etc…

The cool thing about technology trends and curves, is that everyone is staring at them.

If I was a software application architect here’s what I would observe. The biggest cost for an application from a hardware perspective is the storage, and biggest cost factor is the disk drives.

Today with RAID-5, the picture looks like this:

The red controllers are provided by Traditional Legacy Array vendors, and there is a small increase in the total capacity to get excellent resiliency (the pink area).

But with bigger SATA drives, and the absence of a high performance RAID-6, the world looks like this:

Ouch. The total cost of the solution just jumped up! You get cheaper disk drives, but you need more of them!

Now if I am an application architect, I think to myself, you know that RAID stuff can be done in software, and I can do the mirroring at the application level, and the server vendors sell really cheap disk drives, so maybe the picture can really be made to look like this:

Each server has a pool of disk drives, and mirroring is done in the host to obtain some degree of storage resiliency. Yes the costs are greater than RAID-5, but it’s not clear they are that much higher than RAID-10 … And you might even believe they were lower than RAID-10 because you’re saving on the dollars you had to give to the storage company.

So what does this have to do with NetApp?

I really believe that RAID-10 mirroring or DAS is an inefficient way to manage disk drives.

I also believe that the popularity of DAS/Host mirroring solutions is a response to the failure of the market leaders to build an array that can preserve the RAID value proposition.

Since RAID-10 and DAS/Host mirroring are a response but the wrong fundamental approach, I believe that architectures, like NetApp’s, that make RAID-6 work will ultimately win out.

And oftentimes in the debate about DAS vs shared storage, we confuse the specific problems of the Traditional Legacy Array vendors, with the right long term technology trend.