Sharing files with users at remote offices over a WAN often creates enormous headaches, due to limited bandwidth and the high latency of the connection, combined with the “chattiness” of file sharing protocols such as NFS and CIFS.
As a result, just making a quick change to, say, a Microsoft Office document, which usually takes mere seconds on a LAN or MAN (metropolitan area network), takes disproportionately longer over a WAN, thereby frustrating users and reducing productivity.
Unfortunately, intuitive remedies, such as improving the connection, usually cost a bunch and may not bring the expected relief. Alternatively, you can dramatically improve responsiveness by bringing copies of the files closer to the users at each remote location, but the effort of keeping versions consistent can easily nullify the benefit.
With that in mind, it’s not surprising that start-ups such as Actona, recently acquired by Cisco Systems Inc., and Tacit Networks Inc. have jumped on the opportunity to develop solutions that, using a complex mix of compression and buffering techniques, expedite file sharing over the WAN. Since last year, Tacit has been offering a Linux-based version of its I-shared appliance, promising to “make the WAN disappear.” Recently, the company added a Windows NAS-based version that maintains the same architecture but adds improvements such as seamless installations in Microsoft shops, native support for CIFS, and the ability to enforce symmetric file sharing across the WAN. After reviewing the new Windows-based I-shared 1.6.1, I have to admit that the appliance fulfills the promise of making remote file sharing fast and easy, and it doesn’t penalize customers with onerous housekeeping tasks.
Ready, set, WAN
To minimize logistics problems, I reviewed the I-shared on Tacit’s premises, on a dedicated test bed mimicking the typical WAN connection between a central office and a remote one. The central office setting included a Windows Server 2003 domain controller with Active Directories and a Windows 2000 file server hosting approximately 300MB of Microsoft Office files.
To simulate remote users, I used two laptops running Windows XP Pro. On the central office side, the I-shared box acted as a server cache; on the remote side, the second box was running the remote client on IBM xServer hardware. Both machines were running the Microsoft NAS software for Windows 2000, but the forthcoming version of I-shared, expected this month, should also support Windows Storage Server 2003.
To simulate the high latency and limited bandwidth typical of a WAN, I used an application developed by the Massachusetts Institute of Technology, Click Modular Router 1.3, running on a dedicated Linux machine. Using the solution, one can emulate WAN conditions by changing just two configuration parameters: Delay_Shaper to set the latency and Bandwidth_Shaper to set the connection speed.
Setup proved straightforward. Configuring the two I-shared appliances was easy, thanks to Tacit’s intuitive menu entries that are seamlessly integrated with the Microsoft NAS administrative GUI. Instructing the I-shared server to inherit and propagate shared folders from my file server entailed simply typing the server name into the configuration menu.
Moving to the remote client, I found those folders were visible. Using normal Windows commands, I pointed to those folders on one laptop and then mapped the other one directly to the file server shares, bypassing the Tacit appliances. In essence, one laptop was accessing the remote file server with the benefit of the Tacit appliances’ acceleration, whereas the other was exposed to the WAN delays.
Unbelievably fast
I began my tests with a latency of 80ms and full T1 speed, a common setting for many remote offices, and progressively reduced the performance of my simulated WAN. Even at T1 speed, the difference in file access time between those two machines was impressive, to the point that I had to double-check my configuration for trivial errors. There weren’t any.
For example, it took 90 seconds to open an Excel file without going through the Tacit appliance but only 3 seconds with it. Saving the file after making a few changes took 50 seconds without the I-shared. With it, the process took a mere 4 seconds.
In addition to speedier Office applications, I noticed significant improvement when copying files between shared directories and local laptop drives. For instance, using I-shared, the time to copy a 1.5MB Zip file dropped from four minutes to a handful of seconds. Of course, with a slower WAN, the performance gap between using and not using I-share widened even more. On a T1, native access is still possible, although irritatingly slow, but on slower connections, I-shared can make the difference between “impossible to cope with” and “workable.”
To prove that point, I set Click Modular Router to a latency of 600ms, definitely a worst-case scenario, and repeated my tests. As expected, although hampered by satellitelike slowness, working with I-shared was still acceptable. Without it, the agonizing delay attached to each action made getting anything done impossible.
For example, without I-shared, the time to open the same spreadsheet jumped from 90 seconds over a T1 to 10 minutes but remained approximately 5 seconds when buffered by the appliance.
Young but ready
At the end of my evaluation, I have mostly praise for Tacit Networks’ I-shared. The solution crushed delays caused by poor remote connectivity to almost unnoticeable proportions. Moreover, the two appliances passed the test of severe connectivity problems: I pulled the plug with edits pending, and they reconnected graciously when the problem was removed.
Tacit I-shared is a work in progress, which means future versions may offer a wider range of features, but what I saw in my evaluation already makes a well-rounded solution that can improve office productivity and can help your users, both local and remote, walk at the same pace.
Mario Apicella is a senior analyst at the InfoWorld Test Center.
