It has happened twice. Flash back to April 2001. It's Storage Networking World (SNW), Palm Desert, and a well-known analyst has just concluded a presentation. Time for Q&A. An audience member grabs the microphone and asks, "So, what should I install? NAS or SAN?" Now fast forward to this year's SNW. An audience member (maybe even the same one) asks the all-too-familiar question, "So, what should I install? NAS or SAN?"
Twice the same question, one year apart. And if that is not deja vu, the answer is also the same: "It depends. If you want to do networked storage, and it's file I/O you're doing, go NAS [network-attached storage]. If it's block I/O, step up to a SAN [storage area network]."
For sure, no one could be faulted for this advice--twice or a hundred times. In fact, it's common advice heard just about anywhere in storage circles. So then, why does the NAS versus SAN question keep coming up? Is it that users really don't know the difference between file and block I/O? Or, could it be that storage users do in fact know the difference, but in the end don't really care about the difference? Or, could it be that what they are really saying is, "Yes, I know the difference, but I don't want really want to know because worrying about such things really just makes life more complex?"
What users are looking for is something simple and cost-effective, and NAS filers are just that. So, they've started implementing them--despite warnings not to--in data centers to support RDBMS-based applications such as SAP, Siebel CRM, Lawson Financials, etc. Why the caution? Simple. Network latency renders a NAS solution too slow for RDBMS environments--particularly in production environments that require even modest amounts of write I/O activity. Thus, the prevailing wisdom is that NAS is for file I/O, SANs for block I/O. Period.
Why are users bucking the conventional NAS/SAN, file/block wisdom? Basically, it boils down to simplicity and cost. Here's what some users have to say:
"I have an IP network and I don't want another network and another network protocol to deal with just for storage," says one user. "I'm going with NAS for networked storage because it uses IP."
"The performance bottleneck is outside the data center," says another. "In here, manageability is more important. It turns out, NAS makes my DBA's life a whole lot simpler."
"The cost of a SAN was three times that of NAS," explains a third user. "For us it was a no-brainer."
It may seem like I'm writing an ad for NAS here, but I'm not. The point I'm trying to make is that users are inclined to go against conventional wisdom and concentrate on the issues that affect their lives the most. And more often than not, complexity, manageability, and cost are the three biggest issues. I believe the NAS market grew significantly last for the simple reason that it provided users with a relatively easy and relatively low-cost way to move from direct-attached storage (DAS) to networked storage.
Was it cheaper? Yes. Was management easier? Yes. Was it better for the applications at hand? Not necessarily. However, many using NAS for RDBMS have turned to dedicated Gigabit Ethernet subnets between storage and database servers to overcome the network latency issue.
This way of thinking about and deploying NAS has led some users to think that NAS can eventually support all their applications--NAS as generic storage. For these adventurous ones, 2002 has been a good year so far. NAS based on the Direct Access File System (DAFS) became available for the first time earlier this year. A marriage of I/O and file-access technologies, DAFS opens up a pipeline between applications and a NAS device, which allows for better performance--particularly for filers in services to RDBMS-based applications.
Not surprisingly, Network Appliance was the first vendor to implement DAFS in conjunction with software designed specifically for IBM DB2, Oracle 8i/9i, and Sybase ASE. In addition and as an alternative to DAFS, Alacritech and Compaq StorageWorks have since teamed up to demonstrate enhanced NAS performance using Alacritech's TCP off-load processing technology on a prototype Compaq's (now Hewlett-Packard/Compaq) next-generation NAS solution.
These announcements open NAS up to a broader set of applications and move the technology technically in the direction of general-purpose networked storage, making the distinction between file and block I/O less relevant.
Don't expect the NAS versus SAN question to go away just yet. In fact, in light of recent announcements and those to come, the question may be more pressing now than ever before. The only difference is that the answer will depend less on whether the application generates file versus block I/O and more on what IT administrators can afford and how complex or how simple they want their storage environments to be.
John Webster is senior analyst and founder of the Data Mobility Group (www.datamobilitygroup.com), a storage market research and analysis firm, in Nashua, NH. He can be contacted at firstname.lastname@example.org.