Roundtable: "switched on storage arrays" Part 1 of 3. (Tape/Disk/Optical Storage).

Editor-in-chief Mark Ferelli recently joined a panel of experts to discuss some of the underlying factors that are driving the storage industry's move toward switched storage back-ends, or replacing the shared loop buses usually found in disk arrays, mass devices, and some of the scalar systems that use a higher performance switched architecture. Also discussed is which benefits are going to appeal to the end-user in moving from a shared approach to a switched approach. Parts 2 and 3 of this discussion will be printed in upcoming issues of Computer Technology Review.

Mark Ferelli: Thank you, everyone, for being a part of today's roundtable called, "Switched on Storage Arrays." First, we'll do a quick round up of who the players are.

We have here, Arun Taneja who is the founder, president and senior analyst with The Taneja Group, and who's been covering storage related issues for quite some time. He's on his own venture now and previously he was with Enterprise Storage Group and he spent time working in the marketing departments for such players as Andataco and the very challenging Sun Microsystems.

We also have Brian Reed who's vice president of business and market development for Vixel Corporation, who has to execute Vixel's strategic business relationships and identify new markets and that's one of life's little challenges.

Brian Reed: Well, thank you, Mark.

Mark Ferelli: Thank you for being here, Brian, and special thanks to Vixel Corporation for coordinating this event.

Chris Bennett is with us today and Chris is director of platform and system planning in the product marketing group at Network Appliance, and is responsible for development and positioning for Net App's product and system strategies. And, before that, he was a Hewlett-Packard veteran.

Chris Bennett: Glad to be here.

Mark Ferelli: We have James Myers, who's the strategic product marketing manager with Hewlett-Packard Network Storage Solutions and has been there since 2000 in the Enterprise Storage Array Division Marketing Group out in Colorado Springs.

We also have Bob Rumer, the director of strategic marketing for the Datacom Products Division at Vitesse Semiconductor. Bob handles Datacom products and is involved with storage IC product definition and marketing. Product definition and marketing--that's a heavy load, Bob.

Bob Rumer: Well, it's all tied together.

Mark Ferelli: You bet. From LSI Logic we have Mark Nossokoff, who's involved with strategic planning, who's a 17-year veteran, and a senior member of the Storage System strategic planning staff over at LSI, keeping an eye on emerging technologies, entry level solutions, driving closures... and he's also played in the engineering, marketing, and product management positions. Welcome, Mark.

Mark Nossokoff: Thank you, Mark.

Mark Ferelli: And finally, last but by no means least, Jim Beckman, who's director of product marketing at Hitachi Data Systems. He's responsible for the Lightning Series, which is absolutely ubiquitous out in the industry and has played a role in the launch of the 9900 V series, and he started with Hitachi in '89 as a customer engineer, which must have been challenging all by itself and he works under my good friend, Hu Yoshida. Welcome, Jim.

Jim Beckman: Thank you very much. I appreciate it.

Mark Ferelli: Much appreciate your being involved. Let's push right into it, folks, if you don't mind. The first thing I'd like to explore is some of the industry trends that you see as driving the need for a switched storage architecture. And there are a number of ways to play this one. James, I wonder if you'd mind rolling this off. What are the trends that you see as going on?

James Myers: Yes, thank you...The need for higher performance, especially in open systems in the arrays is becoming pretty paramount these days, mainly because most vendors are putting a much larger number of hard-disk drives behind their controllers and especially when these harddisk drives become fatter--that is much higher capacities--now we are looking at how long before we'll see 300 gigabyte-sized drives here, in the near future.

So, the switches really give us at least a 2x performance boost because we deploy them to basically break the arbitrated loops in the radial individual loops and it really helps to eliminate contention.

The second one would be the fact that customers--most of their applications now, even in the mid-range arena besides the high-end enterprise arrays--are demanding 24 by forever reliability, availability, serviceability and recoverability. So, those are the two key ones.

Mark Ferelli: Twenty-four by forever. Anyone who remembers things like backup windows, please raise your hand. Okay, Brian, do you see this as much the same?

Brian Reed: Yes, we at Vixel have been working in the last 18 months to develop and invent these storage switches that enable system vendors to convert from the shared to switched architecture. The reasons why we see the system vendors changing to switched-based architectures are, first of all, the concept of never being down and enhanced RAS capabilities. With shared architectures, you can't isolate problems all the way down to individual drive level and switch-based architectures allow these to do that.

And the second thing we see is the pure economics of services. The fact that the service costs today for both the system vendor himself and end-user is staggeringly high, and switched-based architectures drastically reduce the mean time between repair and the mean time to failures. Therefore, both the system vendor and the end-user benefit from that.

Mark Ferelli: Actually, Brian, that's a rather interesting point. We know that it takes about five times the investment and management to handle almost any piece of hardware, but what we don't always hear about are some of the service costs involved. Jim, how did HDS look at this?

Jim Beckman: Well, I think it really relates to the networking of storage. Just as networking has proliferated, the host attachment has increased exponentially and that drives down to the storage device, so that becomes the central point of access and the congestion builds up there in a bus-based architecture.

So what we have found from a switched-based standpoint is that we're actually driving the network down to the drive level. It provides a lot better access from a performance standpoint, also it allows you to scale the box to a much higher footprint size and do some consolidation in the data center type of environment.

So, storage networking is probably one of the key factors. The other is at the back end, the drive capacity obviously plays into that. You need better access to the drive so that you can maintain consistent performance and then allow the scalability of the storage device in a network environment.

Mark Ferelli: You bet. Mark, what's the view at LSI?

Mark Nossokoff: If you look at some--pretty much any--survey of end-users out there and look at the hot spots where they're going, there's a number of things that come back pretty consistently and one of them is simplified management. It is not only just configuration and deployment, but when there is an issue or a problem, they want to be able to usually diagnose it and get very quickly to the problem resolution.

Isolating the loops down into the individual links to the disk drives can certainly help in problem diagnosis and resolution. Some of the reliability, availability, and serviceability aspects of the switched backend are also important--they have been already commented on.

Also, I'd like to touch on a couple of the other performance areas that the switched-back-end can help with--not only just the mainstream throughput, I/Os per second and bandwidth types of improvement that the switched back-end helps improve, there's also the advanced features in SANs and whether they'd be an outside snapshot copy type of functionality or remote mirroring as well as performance integrated mode when drives are rebuilding. The switched back-end greatly improves performance when some of these operations are going on in the background and can help maintain the throughput and the bandwidth that the applications are requiring.

Mark Ferelli: Tell me, Chris, how about from the NAS side?

Chris Bennett: First off, one of the things that's new about our machines actually started last Fall. They actually can handle both SAN and NAS from the same device. The requirements or the benefits that we're seeing in our space actually apply to both.

The biggest thing I think we've covered is that it enhances the ability to have a higher end system with larger number of spindles, which is really demanded by the increasing performance that customers are expecting from these machines. The other concept here is that it makes it a lot simpler to deploy these. You need fewer HBAs and involvement because you're going to attach more drives to a single HBA environment, so it's simpler to deploy. And then, it also provides a lot better performance, especially in reconstructions going on.

So, to us, it is a natural evolutionary step, moving forward, that suppliers like ourselves and the other folks on the line will naturally adopt to switching. It's just a natural movement forward.

Mark Ferelli: Let's look a moment at some of the advantages of this technology evolution of replacing a shared infrastructure with a switched one.

What do you think are some of the pain points that the architecture addresses both for the end-user and for the solution supplier as well? Let's work on both sides. Jim, would you like to start us off on that?

Jim Beckman: Sure. I think from the end-user standpoint, what we found from moving to a switched architecture--when we had the 7700, which was a bus-based architecture--the average size of that subsystem to a customer was about 1.5 terabyte. As we've gone through the 9960 to the 9900 V series, specifically the 9980, the size of that subsystem has grown to 12 terabytes. The average size of that subsystem is 12-terabyte. So that has allowed the customer to actually consolidate more storage under a single footprint and it has really simplified the management, the diagnosis of problems, the installation, and the network cost requires less host bus adapters, less switch ports. So from a total cost of ownership, it's really decreased the cost of owning the storage device and it's actually simplified the management of that.

From an industry standpoint, from a Hitachi standpoint, it's allowed us to simplify the architecture. It's made it much easier for us to service the product. It has provided us a lot of non-disruptive capabilities so we can actually satisfy the customer requests without having to take the subsystem down and actually allowed us to simplify the performance of the box so that it's much easier to manage. We do not have to do as much placement of storage in specific areas on certain spindles or certain logical units. It is a lot easier to just go ahead and throw our workload on the box, and because it has such great performance and it is kind of equalized throughout the subsystem.

Mark Ferelli: Now I understand. James, how about the HP perspective? Do you get the same things from your end users?

James Myers: Yes. The Enterprise Virtual Array product where we did deploy the Vixel loop switches in our particular product allowed us to actually improve performance substantially. We are actually a member of the Storage Performance Council, as well, and we were able to demonstrate the highest I/Os per second for single controller pair of all those members with the back-end Vixel loop switches because, basically, we got about 7000 more stored I/Os. So you can see that it really is a real world performance improvement.

Another thing that's very key to the end-user is, again, lowering their cost and by putting more hard drives behind controllers and getting a higher density, it really does bring down cost because we all know the controllers are pretty expensive parts of the component solution.

And probably the key thing that the switches have really brought to the market now from a user standpoint is the ability to feel safe about taking both service actions and potential growth or capacity upgrade expansion in real time at the most prime time of the day. In other words, the switches reduced the number of steps in the process to replace or to, for example, add a shelf of additional drives. So that's very key.

On the vendor side, we have seen a remarkably improved manufacturing, testing, and quality and, again, from the service standpoint, we don't have to have our service engineers out in the field as much at a customer location because these are much quicker isolations and replacement actions.

Mark Ferelli: Which saves you a lot of overhead cost right there?

James Myers: Absolutely.

Mark Ferelli: Chris, what's Net App's experience?

Chris Bennett: Well, I think the big one that we're seeing so far is the ability to much more rapidly diagnose issues. When you can do isolation to an individual drive it makes it easier. Loop visibility issues are rare, but when they occur they can be problematic to diagnose. I think that this makes it simpler to diagnose much faster and I guess the correlator a or to that is the ability to diagnose with a much less sophisticated person doing the diagnosis, so that reduces the training expectations of people actually doing the support work.

I think the other key benefit that we're seeing is that it makes it easier to deploy systems and add capacity on the fly because the configuration rules are substantially more flexible with the switched back-end.

Mark Ferelli: Mark, you work directly with the solution providers in that space. What is the pain point that this will relieve from your point of view?

Mark Nossokoff: Yes, an awful lot of the pain points that we're seeing and getting feedback on from the solution providers we deal with are problem diagnosis associated with arbitrated loop, and the ability to more quickly figure out where an issue might be during configuration or set up time. And if somebody has connected something wrong--understanding where that is and being able to quickly fix the user error there, as well as the diagnosis of the problem after it's been installed--should one arise.

So the lower costs associated with supporting the customer and then with the field service call of the end user, as well as alleviating the perception that there are issues with the system that can be quickly diagnosed and quickly solved.

www.vixel.com