EMC CTO Mark Lewis Unplugged
When I asked Mark Lewis, EMC's chief technology officer, why an IT shop should make EMC its storage partner, he challenged me to find examples in this industry's history of a market-leading pure play like EMC losing to a company with multiple product lines. Yet Lewis insists that his company's goal is to not be thought about in the IT organization anymore than a home owner thinks about where the electricity comes from.
ZDNet: Given that a storage strategy must be integral to an IT strategy, let's step away from the storage plan for a second and move a notch up. If you woke up tomorrow as the CTO of a large enterprise that's typical of ZDNet's audience, what long term IT strategy or framework would you be recommending to the CIO or the CEO of your company?
Lewis: The framework for any IT strategy is that you have to look at the major characteristics of what you want to buy and how you want acquire that and what is the long term setup. Five years from now, I will be buying only five things. I'll buy connectivity in the sense of a desktop, a laptop or a personal digital assistant. Then I'll be buying a network infrastructure, some form of computer, some applications, and storage. Those are the five things. Strategically, that's the way I look at the world. You have to find heterogeneous ways to do those five things.
ZDNet: Thinking about those five things --- connectivity (access), networking, computer, applications, and storage --- if you're going into the CIO or CEO's office, what are the priorities? What is your architectural approach? Is it utility computing? Is it browser/thin client-based?
Lewis: It's very much standards-based computing. The first thing you want is to be able to leverage across all of your implementations. Standards can be used to obtain pricing. You don't want to be locked either within or across categories. So you're going to look for what is going to be the standard and you're going to leverage that. You need to look at what are the best-in-class and best-in -reed methodologies for doing things.
This is very different from asking "what is the best technology?" I'm talking about what is the best way. It's kind of like using [Internet Protocol] for networking. Are there arguably better ways to network than IP? Yes. But, what's going to be the standard? What is the standard? You really want to leverage that because that's ultimately what is going to bring your costs in line.
In each of those five areas, you should be trying to deploy once, use many. You should be asking if you can standardize on a database to deploy once across all your applications. Can you standardize on one network over which you can deploy all your applications? Can you standardize on Web services? It goes back to using standards to build your IT infrastructure.
People say there's no value left in IT and that it's all a commodity. I don't believe that. But I believe that you have to buy the best of each infrastructure. When I talked to EMC CIO Dave Ellard, I asked him how many business critical applications he had. I expected the answer to be a few. But he responded with 300! We have 300 applications or more at EMC that run our business. This is e-mail, messaging, collaboration software, Web services, CRM, order tracking, manufacturing systems, etc. We're not going to deploy 300 different networks. We're not going to deploy 300 storage systems. We're not going to deploy 300 policies for archiving information. We're not going to deploy 300 of anything other than 300 applications. Our goal is, can we get our compute infrastructure to one. Can we get our network infrastructure to one? Can we get our storage infrastructure to one? Can we get our access infrastructure down to commonality?
As soon as you go from one to two servers, you're going to need a storage strategy. | ||||
ZDNet: That's a lot of talk about setting standards across an organization. Isn't it equally important to focus on industry standards as well?
Lewis: Absolutely. It's both. Industry standards are needed at major breakpoints, like between servers and storage, servers and networking, access devices and applications. Standards are used in a lot of different places, but they're particularly honed across those breakpoints. Let's say you're going to have a storage infrastructure. You have to take those 300 applications and you have to give one person or one organization responsibility for storing the information. You have a facilities group that runs your building because that's the most economical way to do it. We all don't take out our own trash at night. We all don't change our light bulbs. That's because we built an efficient infrastructure. Obviously, we had to anoint someone with the charge of facilities or we wouldn't get the light bulbs changed. We'd all be running to the stock room and changing light bulbs. So, organizationally, it has to come up front too. The reason you have a network is partially because you have a networks group. That's how it happens. You generally have a desktop services group. They've organized in ways to make themselves very efficient and those natural breakpoints of the business organization will actually be the same breakpoints where EMC has to sell. We have to sell how people want to buy.
ZDNet: Given the IT strategy you just described, how many of the companies out there need a strategy like that? Who are they? Who needs a storage strategy?
Lewis: It's moved dramatically from the enterprise and global 2000 that needed a storage strategy five years ago all the way down into the small and medium business today. Storage becomes a play there the same way as networking became a play. Customers are asking why they're buying storage every time they buy a computer or an application. Why is that they're buying storage for all the wrong reasons?
Also, every public company will need a very specific information management strategy. Once you have to comply with the Sarbanes-Oxley Act, you'll need an information management strategy. Now you're down into the small business category. We started talking about network storage pricing at the entry level around $7,000. So you're not talking about a very big purchase here before you're going to want a storage strategy.
My take is that as soon as you go from one to two servers, you're going to need a storage strategy. When you move beyond one server to your second server, you better figure out how your going to connect the information between the two servers. You don't want to use the two phrases [information strategy and storage strategy] interchangeably, but you kind of need to because you need an information strategy in the sense of information retention, information archiving, information backup and recovery. All of that is what's causing this technology to be driven further down towards smaller businesses. If you're a dentist's office, you need an information strategy to manage the client-patient confidentiality and to manage the protection of those records. What if your office burns down? It seems kind of simple. But that's the problem we have today. If you had your own last will and testament and all these other things you keep copies of; you have your lawyer's office and you have a safe deposit box. Yet, you probably have all your kids' photos sitting on your hard drive in your house and if it burns down, you would be really upset to lose all that.
ZDNet: So, do you have a bunch of CD-ROMs in a safe deposit box?
Lewis: I'm as bad as they come. Actually, we're working on a product called Centera that enables Telco's to provide backup services and archive retention over your DSL link. So, you can have, for $5.95 per month, the ability to back up all the files on your PC. I think it's absolutely the biggest market around. It won't be huge in storage because it won't require a ton of storage, but it's a very important market.
ZDNet: Media management seems to be a major issue there. Searching videos. Searching for still images.
Lewis: The problem with backup isn't backup. What we always tell people is don't worry about backup. Worry about recovery and recovery times and the hassle of management. Disk drives cost nothing. Storage and raw capacity cost is essentially free. Every time I buy a new PC, I copy the entire contents of the old PC onto the new one. Why? Because I can! I might need something. Do I even bother to filter through those files and delete the stuff I don't need? Nope. I just move them all over. That's what we'll do in backup. We'll just store every picture you've ever taken.
ZDNet: Going back to the companies that really need a storage strategy, your saying that the real test is when you go beyond one server, it's time for a storage strategy.
Lewis: Yes. Once you go beyond one server. Beyond that, in a couple of years, every individual will be thinking about a storage strategy because the amount of personal information - your tax records and everything else that you're going to keep - you're going to want a strategy for retrieving all that digital stuff.
ZDNet:What is a storage strategy? I know what an IT strategy is. You've defined that. You talked about who needs a storage strategy. But what is one?
Lewis: A storage strategy is a set of high-level policies that you want to place on groups of information about how it is retained, protected, and accessed. In our vision, a larger company is going to classify data into buckets. Not 25 or 30 buckets, but three to five buckets. The reason you bucket in the first place is for cost. It's for managing a hierarchy of costs. That's why you don't need a billion buckets. You're not talking about a major difference in cost but you will bucket the information. This cuts across all your applications.
If you have e-mail and voice mail and other things and you're complying with SEC regulation 1784 on broker-dealer transactions, you have to retain all of those transactions for a specified period of time. So, you have to ascribe a policy to that information and deal with it. What's different about future storage strategies is that they're applied across applications. Right now, you go into Word for Windows and you can save a copy or you can do something to archive or show information. It's driving your information retention policy based on the application. That's the wrong way to go--building a separate network for every application. That's why we're going to take information and separate it from the application and drive a different behavior.
ZDNet:What about at the hardware and software level? What should ZDNet's audience be thinking about as they attempt to plot a course?
Building a separate network for every application--that's the wrong way to go. | ||||
ZDNet:Are you saying that you have offerings that are optimized for each of those different ways that information is stored?
Lewis: Exactly. So databases want to access blocks. They come in and say 'I want to talk on this volume at this place and I want these two blocks of information.' In order for databases to be fast, they look up random blocks of information all the time. So, there's a type of access called block-level access that we do with storage area networks (SANs) and it's very very fast. There's very little latency and it's high performance and it allows databases to run very very fast.
ZDNet:Does that mean that if you're coming up with a storage strategy, and you are doing database work, that you want to focus on SAN-based solutions?
Lewis: If you're a database guy, you're going to want to deploy a Fibre Channel-based SAN or something else that uses block access storage because, traditionally, a SAN is your primary medium for storing data. It's going to give you the performance. We can build reliability and availability into any system. When making choices, it's really about the data.
Now, you've heard of network attached storage (NAS). It uses protocols like NFS and CIFS. That's typically what we use on file based storage --- otherwise known as file shares ---- here today. Those all run on the principle of file-storing within NAS. So you're storing the entire file as a name and you're asking for files by name. It allows you to share information better. The advantages of NAS are that it's easier to use and it's more collaborative. The disadvantage is that you're reading in the whole file. It's not as granular. It's not as fast. But it's a different need.
The third way we store information is as objects. Objects you can think of as unstructured data. Databases consist of structured data, which means relational records that are usually fairly dynamic and that have highly relational characteristics. Unstructured data is a photograph. That's unstructured data where you're storing a big object with a little bit of information around the object. It's usually what we call fixed content. It's a medical image or an email record or a document that's been scanned in. It's not relational, but you still want it to be a record. We have Content Addressable Storage (CAS), which fits into that market place for storing those records. That's a market that until a couple of years ago, didn't really exist. It was done in a variety of ways. We created a product area that's optimized just around storing this type of data and we now expect that over the next three years, up to 75 percent of data will fall into this fixed content category.
ZDNet:What is the ideal hardware strategy for fixed content?
Lewis: This is NAS and what this actually uses is almost exclusively a software-based product. It uses 1U [link: http://techupdate.zdnet.com/techupdate/stories/main/0,14179,2882078,00.html ] thin servers with ATA drives clustered together. So it's very expensive, off-the-shelf hardware with some software to manage the information. This product is lowest in terms of raw performance, but it also strives to be very low cost , zero maintenance, and zero administration. This is the product that would backup your home PC with adjunct software. So, for example, T-Mobile in Europe sells a service that lets you backup information into this product and it stores that backup as a fixed record.
ZDNet:So, in reality, to the extent that you describe the way the information is stored as a record, the software you describe is actually a database.
Lewis: I don't want to step into a territory where people start thinking that EMC is competing with Oracle. It's an unstructured database, which is a bit of an oxymoron. It's a database for unstructured information. But it's quite different from what you would want to put into an Oracle system.
ZDNet:When I say database, I mean, you're providing a user interface to look up that information that's easier to use than the file handle that one might see in a directory.
Lewis: Correct. The best analogy I've seen about this product is the valet car parker. We create the content address based on the information so it can't be changed and that's done through a mathematical algorithm. So, we'll read in an address to get your car back in the same way that a valet knows where to find your car when you give him your stub. The address is 27 characters long and the application keeps the address. So, if you have the right 27 character ID, you'll get the right information. Simple as that. We know where the car is parked. We can replicate. We can backup. We can do everything in the background and all you expect is that you go to the server, hand it a token, and you get the information.
ZDNet:Let's go back to where you discussed the organizational principles and breakpoints. You described how you have a network group, a building facilities group, and so on, and someone would be in charge of all these groups. Are you saying that there will be a storage group and that someone --- presumably a Chief Storage Officer --- would be in charge of that? Does storage merit that sort of organizational attention?
Lewis: Most major organizations today--and at the very least, all of the Fortune 500--have a storage organization. In many cases, the head of that organization is a director if not a vice president who reports to the CIO. Storage continues to move downward into smaller and smaller organizations. So, we continue to lower the hurdle
Think about how networking worked out. For a while, the biggest companies had networks because they were the most complex and they went to central infrastructures, while smaller companies still handed floppies to each other. But now you see that everyone must have a network. Even if you're a small company, there's somebody whose job it is to look after the network. In small companies, somebody-- even if it isn't a full time person-- is going to be tasked with looking after the company's information.
Even in a small business, the people who run the company will tell you that there are only two things of any real value that can't be replaced: the people and the information. If you lose your people or you lose your data, you lose your business.
ZDNet:What benefits can a company look forward to after having invested in a storage strategy?
Lewis: The first one is regulatory. More and more companies are being faced with information retention regulations. If you're in finance, healthcare, telecommunications or a publicly traded company, you will have regulations that require you to retain certain types of information for certain periods of time. So, the first benefit is simply compliance.
The second benefit is business continuity. You want to have some form of a business continuity plan that deals with data corruption, data loss on site, loss of entire site to some disaster, and so on. Whatever degree you want to ensure the continuity of your business, you need a plan that addresses the all important question of 'how long can I afford to be down?' Every Web site you have should have some kind of disaster-tolerant plan that describes how long the company can afford to have that site down. Then, you have to throw varying degrees of money at it to match the protection you're looking for. That's how you drive the business continuity.
The third thing-- and the thing that's most tied into all of that and that's going to make or break you-- is your overall storage management strategy. With the cost of storage essentially being free, the problem is that all of us users feel that it's our God-given right to store everything we've ever created. What ticks me off are e-mail restrictions. Why should I have to go cull through my e-mail once a week? I'd rather just give the IT guys another disk drive if capacity is the problem. But the problem isn't the storage cost. It's the management cost around storage. They'll say, "Oh yeah. You want us to store it. But then you probably want me to back it up and then you want me to archive it and then you want me to give you fast performance and all those things with it. Yeah, sure. You want to give me the drive? That's the easy part. I have to put it my SAN. I have to administer it. I have to provision it. If it fails, I have to call service." These are all the things they hate.
For every dollar you spend on a drive, you're spending about seven dollars in administrative costs to keep track of the dollar. | ||||
ZDNet:How do think that compares to the other areas that you've outlined, like networking. Usually, it's a no-brainer for someone to tell you how much network capacity they have.
Lewis: Let me give you the adoption curve. Networking is the furthest along. It's the most mature. What did it evolve out of? It evolved out of taking the intelligence out of central server and putting smarts into the network. What's the second most farthest along? The application itself. They're getting pretty good within the applications. They're getting a lot of tools to look at deployment of individual applications. Cross-application stuff, we're not there yet. Next, I believe, is storage. We're starting to build provisioning tools and storage resource management tools. Storage utilities have been around for five years and now compute utilities are becoming the rage. Computers figured it out from storage. It's not the other way around. Everything is looking at utility computing and saying 'look at this great trend.' Well, we've been doing this for a long time. One of the things we done, like with networking, is that we're moving more and more intelligence out of the server. The thing that's going to be left with the least amount of intelligence is the server. Think about it. Back in the days of the mainframe, we had the mainframe and everything else was called a peripheral. Well, the whole model of computing is imploding .
Scott McNealy says the network is the computer? Nah. The computer is the network. It's going to go exactly the opposite of how Scott predicts. The least important thing in your IT infrastructure is going to be your computer. You're going to have an important network to get your information. You will have an information store that is your information. All your intelligence is going to exist within your connectivity framework. That's where all the smarts have to be to deliver the data effectively and securely. That all occurs in the fabric. What are all those bumps in the middle? Sure, they're sitting there taking this information and processing it so it can be consumed. But it if you looked at it from the beginning, first you had really smart computers. Then you had client/server so you had these two smart things connected by dumb wires. Now what's happening is the wires are getting really, really smart and the ratios will change all the way across the industry.
ZDNet:You identified business issues that a storage strategy addresses, such as regulation, business continuity and storage management. I didn't hear you mention the term 'cost.' That's usually an issue that routinely comes up as a hot button amongst ZDNet's audience--driving down the total cost of ownership of the infrastructure, for example.
Lewis: Cost is driven by each one of those. Storage management is the biggest driver of raw cost. When a company is looking at a desktop or laptop strategy, you're also looking for consistency and maintenance and other things. Raw cost will exist at a competitive rate across all storage. You have to determine your business requirements first because you'll have to pay for those no matter what. You need to understand that it's about running your business. You can't start from a cost perspective up front. You have to start from a business perspective or you'll never invest in protecting your information. If it was all about cost, I wouldn't spend the two hours every month creating the DVDs and putting them in the fire safe. It's not cost effective until I have a fire. So you plan for that. Management is all about taking cost out. That's where the TCO model works. Think about any other form of scaling. The reason you have the storage and the infrastructure is because you're dealing with an equation where you're putting in a base fixed cost and your variable is the cost of buying additional hardware. That's because you've figured out the business policies, you're leveraging a scalable management scheme, and you're leveraging all that infrastructure.
IT departments should be looking at the serviceable life and TCO versus acquisition costs. Now, acquisition costs still need to be competitive because nobody is going to take that completely out of their mind. This is not an excuse to say that this is going to be a lot more expensive. But further and further down into the smaller and medium businesses, they're starting to say 'Well, wait a minute, I could go out and buy that cassette answering machine recorder, but I'll probably use voice mail because it's a service, it's there, it's more dependable, and I don't have to replace it all the time.'
ZDNet:Looking forward at some of the broader issues to which storage is somehow connected, I want to throw some words and phrases at you to get your gut response.
Lewis: Shoot
ZDNet:Spam
Lewis: (laughing) You know, EMC thrives on content! We have some interesting things that fall into storage with not just spam, but with content replication in general. For example, we've optimized our object-based storage like this. Suppose I can read in your business card as a content address. I also know the exact contents by the address. So, if you have a PowerPoint file and you send it to me and then I go to back it up or store it, the system knows for a fact that it's the same file that was sent to 99 other people. So, we store it once instead of 100 times. And then, we compress it. To some extent, that helps mitigate capacity issues associated with spam. But a real problem now is with regulations. The worst thing is spam coming into a brokerage house because they're required by law to keep everything, because they don't want the five or ten dead minutes on the tape where someone says "oh, that really wasn't important." The agencies need to be able to see everything. In order to get the whole truth, you have to be able to show that every record was searched and here's what was found, not just the ones that someone thought to be important. The controlling of spam is very important because we're saving it and replicating it. But from a storage perspective we're also doing what we can to compress it and minimize the capacity requirements to deal with it.
ZDNet:Rich media devices like wireless camera phones or MP3 players that generate a lot of content. When I think about the explosion of rich content, the next logical question is, "where is all that content going to be stored?"
No integrator has ever won a battle against a focused best-of-breed competitor. | ||||
So, we've been talking about how there needs to be an information architecture to your data. We believe that it has swung back and forth between centralized content and distributed content. For example, it went from mainframe computing to client/server computing. Then it went back to mainframe and then back to client/server. My favorite was watching HP with their latest announcement of the centralized server thing and I was like "It's a terminal! They've gone completely back to terminals." I couldn't believe it. I actually don't believe that that's the right way to go. What I believe is that everything is going to have storage built-in and the reason you put storage closest to the device that needs it is for speed, pure and simple. Also, sometimes, you might not be connected. But more and more, wherever you are, you're connected. But, sometimes, you might not be able to be connected. Speed and reliability of the connection are why you have data distributed, and we want to maintain that. We think that's the right place.
I love my TIVO in my house. It keeps a big cache of what I want to watch. It's just a cache, right? The difference is that the information also needs to be centralized. Why do you centralize? You centralize for protection of the information and control of the information. When I bought this new phone, I should have been able to dial my number, say my name, and have my phone book and all of my other personal settings automatically drop back into my phone from the phone company. Verizon is a big customer of ours and I tell them that all the time that they should be storing my contact list and my address book for each new phone. If I get a new PDA or I lose my phone in Europe, I should be able to walk over to your phone and do the same thing so that your phone becomes my phone.
ZDNet:Well there's always the SIM card, right?
Lewis: Right, that's hardware. That's exactly what I was talking about before: You need storage local to the device, but you also need it centralized. What happens if I lose the phone? The SIM card was in it. So, this business is going to be more and more about the software that can sit in the phone and that can sit in the centralized database, that can manage file and content distribution and that can handle that in a very seamless way.
To be honest, our goal is to not be thought about in the IT organization. It is to be disrespected, if you will. The last thing you think about in your home is where you are going to get your electricity from. You haven't thought about it for years. It's just there. It's an efficient service that's being dealt with behind the scenes and that's what we think ultimately needs to happen to all this information. You'll store more if you can manage it less. You'll get more email storage if your e-mail administrator doesn't have to deal with the complexities of managing the storage. So, EMC's job is how do we get you to store more. We have to make the management tools better so it costs you less to manage.
ZDNet:It seems like the next evolutionary step is to outsource storage altogether? Is that the case?
Lewis: The reason you just don't centralize all the storage is, that there's one law we cannot repeal: the speed of light. It's the biggest problem we have in IT today. Everything else is moving along at Moore's Law here. Yet it's still the exact amount of latency from New York to L.A. today as it was 10, 20, or 30 years ago. That's why cache content and content distribution works so effectively. Because when I punch that phone number and I say my voice in there, I need local memory to deal with the processing capability.
How does this get to the outsourcing question? The same thing is true of databases. While storage is an infrastructure, the economic payback for most large companies comes very quickly by keeping the infrastructure insourced with their applications. If they outsource the whole thing, yes. But you will see very little in the way of outsourced storage with insourced applications. The distance itself of putting storage separate from applications does not work. While you're going to have a network of storage and network of computers, we believe those two networks will be in the same room physically.
ZDNet:So, by virtue of the way somebody like salesforce.com takes over your application, they take over your storage too?
Lewis: Exactly. So, what I do believe is a strong trend is business process outsourcing, not IT outsourcing. I'm a fan of insourcing because I think IT is a competitive weapon. Will we outsource at EMC? Sure, we outsource a business process. We do not say "go run my computer for me." Getting someone to run your computer does not give you a strategic edge. But with companies like ADP and SaleForce.com, these things are good for smaller companies that don't have a core competency there. You don't have a core competency in payroll? Great. Go to ADP. In that case, we just sell to ADP because they're going to build an IT infrastructure that's optimized around all their clients. Our sell-to person is just a different person.
ZDNet:So, if I'm an IT guy, why should EMC be my storage partner?
Lewis: I'm a student of technology history as much as technology futures. I think you can use technology history to predict the future. If you go through maturation cycles, it's not that innovation stops in these areas. But that the most effective innovation over time is where two or three players invest enough in a particular category to be leaders in that category. The market stays price competitive because of competition, but the market also advances even faster because you have the R&D focus of a couple companies with a large marketshare. Pick your area. It almost always happens. Pick processors. Pick disk drives. Pick networking. What is always the end game? So, why is EMC is going to be successful? Who else is focusing on storage?
I came from HP. To me, it wasn't about who had smarter engineers or who had better R&D. You could see the writing on the wall. You could say, "OK, let's see who is going to wake up in the morning and think about storage and how to make people's information lives better?" Then, you could ask "Who's going to wake up today thinking about printers, computers, services, storage, and all these other things." [HP CEO] Carly Fiorina doesn't lose her bonus if the storage business goes away. If printing and imaging goes away, well that's a different story and that would impact her in a big way. This entire company runs on storage.
Looking back over history, who consistently wins out? Did Digital win the processor wars? Do you think Sun SPARC is going to win the processor wars? No integrator has ever won a battle against a focused best-of-breed competitor. Not ever! It doesn't happen. Slowly, IBM has gotten out of every best-of-breed component. They've tried to get back in and they're probably the best at holding on. I will give them that. But take OS/2. Did it make it?
It's always been a best-of-breed focused player that at the end of the day has succeeded in the marketplace.