SAN/NAS Q&A with Steve Duplessie

Posted on July 01, 2000


By Steve Duplessie

Steve Duplessie - Enterprise Storage Group
Click here to enlarge image

The following Q&A session was excerpted from an online chat session hosted by searchstorage.com, a Web-based search engine site for storage-related articles and information. For more information, visit www.searchstorage.com. Questions were posted by end users.


Is there a best-of-breed SAN solution? If not, when will there be one?

No. There are best-of-breed SAN vendors who know what pieces work with what to provide a solution. All of the vendors have strengths and weaknesses, and no one is a catchall. There will never be a single winner. What will happen, like with IP, is ultimately standards will be adhered to, and the best technology for the given task will be chosen-and we'll know it works with all the other pieces.

Some vendors have built special labs to deal with interoperability problems. Do you think this will really make a difference?

No. Vendor interoperability labs are designed to make sure that a vendor's products work with other vendors' products, but not necessarily with competing solutions. For example, Compaq isn't worrying about working with EMC; they're worrying about working with Vixel, Brocade, Gadzoox, etc. That isn't the problem. The problem is: How do multi-STORAGE shops interoperate?

What do you mean by multi-storage shops, and what are the specific problems we should look for?

By multi-storage shops I mean sites with storage subsystems from different vendors-for example, an existing Compaq StorageWorks customer that is considering bringing in an MTI box. The goal here is virtualization of the total storage environment, because array #1's management software never considered a foreign array in the equation.

How will multi-storage shops interoperate in the future as the pressure to do so mounts from customers?

Watch the SAN appliance space. These are vendors that put volume management into black boxes to control the SAN. With appliances, users will be able to configure and use any combination of disk arrays with any kinds of hosts-securely. The SAN appliance vendors are doing what the switch vendors should have done long ago. As soon as SAN appliances are ready for prime time, 90% of the SAN problem areas go away.

Is there any way around the interoperability problems? Is the only solution to standardize on one vendor?

Today that really is the only viable way. Within only a few months, however, you'll be able to mix and match vendors freely. There are interoperability standards proposals, like OSFI for switches, that will work. The problem has been vendors' reluctance to conform.

The major switch vendors are working together to enhance interoperability between brands. How is that effort going?

Switch-to-switch E port interoperability was demonstrated at N+I. Common zoning is a problem that won't be fixed till year end, however. There are ways around that though, such as using more discrete partitioning in either the arrays, SAN appliance, or hosts.

Is there some way to take a building-block approach to implementing a SAN-to ease into it and avoid some of the interoperability issues?

Sure, stick with a pre-packaged "SAN-in-a-box" environment, a la Compaq. These smaller configurations are all pre-tested and pre-configured and, most importantly, they work. If you have existing hosts/storage you want to configure into a SAN, it gets more complex, but it can be done.

With all the interoperability problems of SANs, what's the compelling reason to deploy one today?

Because we have lots of hosts that need direct physical access to lots of disks, most likely in a clustered configuration.

Is it possible to implement a multivendor SAN, or is it better to stick with one vendor?

Sure it's possible. It's just not as easy. Everyone supports the big intermediary vendors, like Brocade, Gadzoox, Vixel, and Ancor. The problem isn't making things work, it's having 25 different management pieces. Again, the SAN appliance folks aim to make this reality, and from what I've seen in the early stages, it is very promising.

Is there any way to design a SAN with no points of failure?

Absolutely. As a matter of fact, don't configure it with any single points of failure. Use two of everything in the data path.

A number of storage vendors are talking about point-in-time recovery capability. Can you explain what that is and its relevance to storage management?

Think of it as "snapshots". Disk subsystem vendors talk about point-in-time images as the ability to create a volume replicate (EMC's Timefinder, for example). When a problem occurs, you can go back in time to a known good state instantly, instead of restoring from tape.

How do you select a SAN vendor? Does any vendor offer a complete solution?

Some vendors offer end-to-end SAN solutions, but the problem is they are only interested in making their own stuff work. If a customer wants to buy Compaq arrays and already has an existing EMC infrastructure, it gets tougher. Look for a competent integrator.

What should a company look at before deciding to use SAN versus NAS?

At a basic level, do you want to share a file or a disk? File = NAS, disk = SAN. Also, where does the application execute? At the desktop = NAS, local = SAN. Of course there are lots of issues to consider, but those are the basics.

What SAN design guidelines are available?

There's a good book called "Building Storage Networks," by Marc Farley. You could also start with our SAN report for the basics: www.enterprisestoragegroup.com

How can we ensure that migration from SAN islands today will not end up in a nightmare when merging them to "clouds" in the future?

You will get Fibre Channel switch interoperability soon. And IP fabrics are going to come on strong, allowing you to tie SAN islands together much easier, both locally and remotely. We are firm believers that storage networks and IP networks are destined to merge.

What "killer apps" does SAN offer?

Backup and clustering.

Is there 'real' SAN management software available in an open systems environment today?

Yes. One example is Vixel's SAN InSite 2000. It manages any device that adheres to the Fibre Alliance MIB, which is pretty much everything.

Will Brocade drive the market in the future?

Brocade has thus far bet the farm on Fibre Channel fabrics. They are clearly the market leader, but what happens when Cisco changes the rules to create fabrics via IP?

Do you see any switch vendor challenging Brocade as the leader for the foreseeable future?

Not until the existing legacy installed base starts to deploy SANs. Brocade owns the OEM space, and it will be almost impossible to unseat them.

Many people believe that Fibre Channel and SANs are the same thing. What's your opinion?

Fibre Channel is an architecture, SAN is a concept. They are pretty much thought of as one and the same, but that may change as IP enters the picture. For now, SAN is how to configure a storage network, and Fibre Channel is the dominant device architecture used.

Can you build a SAN using switched Gigabit Ethernet? Or is Fibre Channel what defines a SAN?

Non-IP shared storage connected to multiple hosts is my definition of a SAN for now. Once vendors figure out how to move block data over IP I'll add Gigabit Ethernet to my definition. If it's block data movement, it's SAN; if it's file based, it's NAS. When storage over IP happens, it will qualify because the hosts will still talk BLOCKS, which will be converted to IP frames.

What is the state of SAN-attached backup and direct-to-tape backup software?

CA announced it, but only for their own world. Legato's Celestra is way up on this front. OEMs and partners will be rolling out server-less backup configurations eminently.

What is a "data mover"?

A product that invokes SCSI extended copy commands to move data directly between disk and tape in a server-less backup environment.

What about a global file system? Will there be a future for the file system in the SAN/storage backend?

Watch for Veritas to come out with the first rev by year end. More importantly, the SAN appliance vendors will help sooner. Global file systems are awfully tough to build.

I saw a presentation by FogDog Sports discussing how they utilize NAS to store both Web content and RDBMS data, in this case on Network Appliance hardware. I'm most familiar with RDBMS data being stored on locally attached disk. Is storing RDBMS data on NAS devices a growing trend?

RDBMS served via the net is cool for distributed Internet database applications, but direct-attached storage for OLTP always provides better performance. Because more companies are adhering to the Internet paradigm, I expect NAS-serviced RDBMS configurations to continue to grow.

Do you believe that systems integrators are up to the task of building SANs?

Good ones, yes, but there are few. Talk to the switch vendors, because they know who has their act together. Integrators are important, because they not only know the switch issues they also know the storage and management issues.

Do you see an increasing role/presence/opportunity for systems integrators in SAN acceptance?

Systems integrators have huge potential with SANs. Users don't know, or don't want to know, how it all works: they just know they need it. Integrators that have solid storage/SAN talent can help a lot of companies, and as such make a lot of money.

Can you provide any guidance on integrating SAN solutions with server clustering?

Traditional clustering has been limited due to the amount of servers we could physically attach to shared storage. SANs effectively eliminate that problem. Homogeneous clustering is relatively easy, but if you combine clusters you have to make absolutely sure that cluster #1 cannot see/access the devices owned by cluster #2.

If I buy a pre-packaged SAN-in-a-box solution (i.e. Compaq), how can I expand on that later? Do you think the SAN-in-a-box vendors will provide for heterogeneous support? When?

Compaq won't rush to support EMC, but that won't matter. What matters is that the switch the OEM uses will support multiple storage array vendors, and somebody will offer you unified management. Pre-canned SANs are a great way to get started, with limited headaches. The trick is to make sure that the components haven't been modified to become "proprietary."

What are the pros and cons of some of the recent IP SAN solutions you've heard about?

I can't give you details (because of non-disclosure agreements) but there is some really cool stuff coming out. There are no real con's (assuming vendors figure out how to move block data over IP), and a lot of pro's (we all know IP, it works, no interoperability issues, etc.) First, vendors will bridge SAN islands via IP fabrics.

Will InfiniBand replace Fibre Channel?

No. I think InfiniBand will replace PCI as an inter-processor bus, but that will probably be the end of it. I don't think we'll see InfiniBand devices, so Fibre Channel has a lot of life left.

As the backbone of a SAN, does Fibre Channel have anything to fear from Ethernet and IP?

Yes. I think that Fibre Channel switched loops are going to continue to dominate, but a combo of IP and Fibre Channel backbones will emerge. As soon as Cisco says so, the world will change.

Will IP fabrics suffer from the same security concerns that IP-based NAS storage does?

Yes. The bet is that the IP vendors will figure it out and standardize quicker than the Fibre Channel vendors.

Are there any issues with implementing an NT cluster with a SAN?

Yes and no. Issue #1 is Microsoft, and the fix is software from vendors such as Veritas or Legato. Use Veritas Volume Manager and most problems will go away.

What about problem solving, which can get complex with different vendors? Are there any monitoring tools to assist in identifying the problems/bottlenecks?

Storage resource management, or SRM, software will dramatically improve your ability to see where your problems are coming from. SRM software is available from vendors such as HighGround.

Can one server write a file to a SAN and another server read from the same file simultaneously, or is it dependant on the file locking provided by the software?

This is dangerous ground. You can let any host you want see any volume you want, then go into the secondary host and designate the volume as read only, but watch out for Murphy's Law. Otherwise you're talking clusters, or software like Tivoli's SANergy to "share" files, although it's not really sharing.

With an all-NT environment running Exchange and no other applications, what do you suggest in regard to NAS?

I think NAS is a great repository for Exchange data. Let's face it, we never throw e-mail out, even though we rarely re-access it. It clogs up our enterprise backup and slows down the network. Putting it on a dedicated NAS device makes a lot of sense.

I've done some reading on SANs and NAS, but it seems the bulk of the solutions are very expensive. Is there an inexpensive solution that implements NTFS security?

There are lots of NAS devices that handle NTFS security. I suppose it's a question of "what is expensive?"

What are the advantages to having a company like StorageNetworks simply manage your storage for you?

I love the whole storage services provider, or SSP, concept. It works because of two fundamental principals: (a) online storage growth is exploding at a ridiculous rate and (b) there is little storage talent left in the world. Those two elements make it awfully compelling for me as the CEO/CIO to say "Hey, let's go back to our core competency, keep our intellectual property (content) to ourselves, and give the boring infrastructure stuff to the professionals."

With the emergence of storage services providers, is there a reason to worry about whether it's a SAN or NAS solution, or about interoperability problems? What are the down sides to an SSP solution?

Conceptually, none. SSPs sell you a utility based on a service level agreement, and if they don't meet it, you don't pay. It's up to them to determine the best way to meet the SLA.

We have a large investment in SCSI storage units and are looking into SANs. What are the most likely steps to use existing hardware and still progress forward with SANs?

Most SCSI subsystems are upgradeable to Fibre Channel. If not, there are plenty of Fibre Channel-to-SCSI bridge products from vendors such as Crossroads, Atto, Chaparrall, etc.

Can you address some objections to SANs-e.g. cost, more storage means more management, and lack of good tape technologies?

I would argue that it costs much more to NOT have a SAN (ongoing management, etc.) There are plenty of great tape solutions. As a matter of fact, backup is SAN's first real killer app. Plus, consolidating to SAN will improve overall efficiencies and let you use more of your disk space than traditional methodologies.

Steve Duplessie is a senior analyst with Enterprise Storage Group, www.enterprisestoragegroup.com.

Comment and Contribute
(Maximum characters: 1200). You have
characters left.