Branch offices. Don't you just hate them? They may be vital for business, but from a network administrator's point of view they cause no end of problems.
Here's one of them: you have several (or hundreds of) branch offices, and the users in each need access to the same corporate data. And, naturally, all the data has to be backed up regularly.
The obvious course of action is to centralize all the data at a single location, allowing branch offices to access it over the WAN. Except this is a terrible solution. Performance over the WAN is bound to be unacceptably slow, and increasing the bandwidth of your WAN links won't help much because this is a latency issue as much as anything else.
Another solution is to replicate the data once a day at each branch office - so each user accesses it locally - synchronizing and backing it up at night. But then you have the problem of multiple out of date copies of any file which is modified, and you need the same storage volume at each branch office as you do at your corporate headquarters to store all the replicated data. Your storage costs will go through the roof.
The WAFS architecture is fairly straightforward. At the corporate data center, a WAFS central server is attached to NAS or a SAN. This central server is connected over the corporate WAN to a device called an edge file gateway at each branch office, and this is connected to the branch office LAN. At each branch office, users see the data they need on their LAN, apparently stored on the edge file gateway just as if it was a local file server. Performance wise, the data is available at LAN-like speeds.
What is actually going on is a little more complex. The central server is responsible for permissions, access controls, data integrity, file management and data protection at the remote sites. The first time a file is accessed at a branch office in a given session a cold hit - it is sent over the WAN to the edge file gateway using various network optimization techniques including compression, data streaming and differencing (sending only the data that has changed from a previously accessed version of the file). Subsequent file accesses (warm hits) come from the edge file gateway's cache, and are supplied over the LAN only, at a much higher speed.
If a particular business process involving certain data sets takes place regularly, it is even possible to pre-populate the cache with the required data beforehand, so the first time the data is requested it is a warm hot from the edge file gateway cache rather than a cold hit delivered over the WAN.
The central server hands out a "license" to a user accessing a given file, so it can only be accessed by one person at a time - eliminating problems caused by having multiple edits of the same file - and changes to the file are sent back over the WAN to the data center, either immediately (write though), or via the edge file gateway (write back).
So does it work? The statistics provided by the vendors are certainly impressive: Noah Breslow, vice president of marketing at South Plainfield, NJ, based Tacit Networks, says the time to open a 5MB file on a branch office over a T1 line with 60ms latency can drop from 122 seconds to 11 seconds for a cold hit, and then 3 seconds for subsequent warm hits. Write speeds fall from 81 seconds to 4 seconds.
Brad O'Neill, senior analyst at Hopkinton, MA based market analysts Taneja Group, says these claims are realistic. "My original question when I looked at WAFS was Is it truly close enough to offering LAN like performance that users won't notice they are using a WAN?' - and my conclusion is that it is."
But, as it turns out, many of the companies that are currently testing these solutions are not primarily interested in file sharing. That's because WAFS offers two value propositions: file collaboration, and centralized backups. And what's driving interest in WAFS, says O'Neill, is the possibility of consolidating backup procedures.
With a WAFS system in place it's possible to eliminate branch office-level backups, with all the associated hardware, storage and support needs because all data is held - and can therefore be backed up - centrally. "If you can remove several thousand dollars of costs per branch each year by consolidating remote backups, then this is very appealing," says O'Neill. Centralized storage and backup of data also makes data protection and compliance with laws such as Sarbanes-Oxley much easier, and the productivity gains from faster file sharing are not to be sniffed at. "The early indications from end users are that ROI rationalizes itself on a per site basis. So even if you just had one or two remote sites, it could be worth implementing."
The main obstacle to the adoption of WAFS is that the vendors concerned are all relatively young: no CIO in their right mind is going to switch something as important as data backup to a new technology from a young company. But in the last few months Cisco has bought Actona and EMC has announced a partnership with Riverbed. Expect more deals in the near future.
It's still early days for WAFS technology, but the likelihood is that, like anything that promises to save money, we'll be hearing plenty more about it in the future. "There is a potentially compelling ROI story here for a huge number of companies," O'Neil concludes. Watch this space.