Just when it seemed as if a truce might be feasible, fighting has flared up again in the battle between the WAN optimisation vendors over just who can - or can't - accelerate the latest AutoCAD format for drawing files.
The vendors - not only Riverbed Technologies and Silver Peak Systems, but also Cisco and others - have been boosting themselves and badmouthing each other ever since it became apparent some months ago that users upgrading from older versions of AutoCAD found their WAN accelerators weren't working as well as they used to.
The problem was the file format: AutoCAD's developers had added compression to its 3D model data, and in the process they were scrambling the files. The result was that each time you did a save, even if few or no changes were made, it looked like a new file - and that caused problems for anything trying to cache or de-duplicate it.
For several of the companies affected, this was a good example of what can happen when programmers implement changes without considering the entire infrastructure in which their software must operate.
To pour oil on troubled waters, AutoCAD's owner Autodesk joined up with Riverbed, which appeared to be the WAN acceleration company most heavily affected - it estimates that around 10 percent of its several thousand customers are in this sector. The pair issued a joint statement, advising users on how to adjust a key AutoCAD setting called ISP (incremental save percentage) to minimise the problem.
On top of that, Autodesk said it was determined to fix the problem completely in its next major release. It said that in future it will take account of its place in the IT ecosystem, and its need to interwork with others.
However, Silver Peak is still telling anyone who'll listen that its WAN acceleration boxes are not affected by the file format issue anyhow, and Cisco and Packeteer (now owned by Blue Coat) have said similar things. It's worth noting that the latter pair use local file caching - this will indeed give better local write performance, but there can be consistency issues.
Meanwhile, Silver Peak claimed its advantage was because most of its rivals use token-based de-duplication, while it uses a more efficient form, which it calls instruction-based de-duplication. (A document on the topic, The Benefits of Byte Level WAN De-duplication, is available online - it requires registration though.)
A key difference, said its product marketing director Jeff Aaron, is that the former is limited to working on blocks, and it cannot detect when the same data is simply shifted along a few bits. By contrast, his company's version looks at data as a stream so it can detect patterns shared by two versions of a file, even if they're in differing parts of those files.
That's really got Riverbed's back up - it says Silver Peak is wrong, and it has done the hardware tests to prove it.
To be specific, it took its own Steelhead WAN acceleration boxes and compared them to similar devices from Blue Coat, Cisco and Silver Peak - the latter two as they claimed to be unaffected by the file format changes, the first as it hadn't involved itself in the AutoCAD fracas, Saldich said.
The tests involved every possible permutation of new and older AutoCAD formats, two different WAN bandwidths (1Mbit/s and 10Mbit/s), an ISP of 0 (full scrambling) or 50 (partial scrambling), and what the test report calls major and minor file changes. With four binary factors and four pairs of devices, that meant Riverbed's technicians ran 64 different tests in total.
The Riverbed results, which have been validated by storage analyst firm Taneja Group and are available online (more registration needed - sorry!), show that all four vendors suffered similarly from the change in file format. With ISP set to 0 they were still able to accelerate the file-save process, presumably by optimising the network and application protocols, but not to reduce the amount of data transferred.
However, by far the biggest gain in save-speed for AutoCAD 2007 came not from different de-duplication technology, but from changing the ISP setting, as Autodesk had already advised users to do.
"We don't dispute they do de-duplication differently, but we're confident it's irrelevant - at the end of the day, we're all substituting a small amount of data for a large one," asserted Alan Saldich, Riverbed's VP of of product marketing and alliances.
He added: "Silver Peak is just trying to get PR - we've proven they're wrong, and caught them with their pants down. We'll publish details of our tests, and we welcome people replicating them."
Looking forward, it seems quite likely that the skirmishes will continue as the vendors all seek to give themselves a leg up in this combative market.
Hopefully though, at least one good thing will have come out of it all. That's the growing realisation that just as networks and storage can no longer operate in isolation, but must take account of their effects on application performance, so must application designers take account of their effects on the rest of the infrastructure.