Although Cisco Systems Inc. claims its WAAS products work fine with the AutoCAD files that have performance issues with Riverbed Steelhead appliances, one Cisco customer performed specific testing and found similar problems while saving AutoCAD 2007/2008 files.
The issue first came to light around six weeks ago when engineering firms using Riverbed and AutoCAD discovered it took longer to save files from remote sites with the latest versions of AutoCAD. The problem started when AutoCAD changed its file drawing format for AutoCAD 2007 and it continued with AutoCAD 2008. Beginning with AutoCAD 2007, the application rescrambles all bits of a file on every save, even if no changes are made to the file. This neutralizes the advantages of data deduplication that most WAFS products use.
Riverbed and several of its rivals have publicly sparred about the causes of the AutoCAD problem and whether it affects all WAN optimization products. AutoCAD has said it plans to change the way ISP zero files work in the next release of the file format to better accommodate WAN optimization systems, but some customers are frustrated by the wait for a solution.
Cisco customers put WAAS to the test
Jeremy Gill, vice president of information technology at Pittsburgh civil engineering firm Michael Baker Corp., had his network engineers conduct specific testing on its data redundancy elimination (DRE) with the help of Cisco service engineers last week. Michael Baker uses Cisco WAAS to optimize its WAN across 30 locations. Although Gill said no problems have been noticed in day-to-day operations, tests showed a sharp discrepancy between optimization on files with an ISP of 50 and an ISP of zero.
Gill's staff performed several tests on a 10.2 MB drawing file over a DS3 to T1 connection. The file was first opened with AutoCAD, and DRE statistics were measured. Those statistics were then cleared, and a second pass was conducted. Finally, the file was changed and saved back with an ISP of 50 as well as an ISP of zero.
On the final tests, the save with an ISP of 50 left 904 KB unoptimized, meaning the system achieved a 90.05% optimization. But the save with an ISP of zero left about 10 MB unoptimized, a 9.43% optimization. "I can't really explain what that says," Gill said. "But in our day-to-day operations we use an ISP of 50, so we haven't seen a difference."
Riverbed executives claim any data deduplication system will have the same issue that their Steelhead appliances have. "It's a very specific use case," said Alan Saldich, Riverbed vice president of marketing. "We can still optimize files with ISP 50 or 100."
Different approaches, same results?Since Riverbed's problems surfaced, competitors Cisco, Silver Peak and Packeteer have insisted they don't have the same issues as Riverbed. Silver Peak recently issued a white paper detailing how its instruction-based method of optimizing WAN traffic gives it an edge over Riverbed's approach of using 16-bit tokens to represent deduplicated chunks of data over the wire. Cisco's DRE also uses the instruction-based approach.
But regardless of how deduplicated data is represented over the wire, the data must first be reduced by recognizing patterns. Riverbed said that makes the instruction-based vs. token-based argument a red herring. Saldich also said Riverbed is planning to release the results of head-to-head testing of its box against competitors' products next week, with process validation by analyst firms.
Packeteer, meanwhile, doesn't perform the same kind of block-level deduplication as the other products and is achieving its results using local file caching.
A Cisco spokesman said the Michael Baker tests reveal that WAAS improved performance, and that improvement would have been greater if the engineering firm could use CIFS optimization. "This particular test was performed on a Novell system, so CIFS optimization was not used on WAAS," said Feng Meng, solutions marketing manager for Cisco WAAS. "Despite that, the first-time download still gave the customer some optimization." Meng said Cisco's internal tests with CIFS optimization from local file caching show the file would be optimized 10%. But he claims caching would still result in a response time savings of 40%.
Riverbed's Saldich said using file caching to save the file only abstracts the file save to the user – it doesn't make it faster. "It's like mailing a FedEx envelope," he said. "Once you drop it in the box, to you, it's done, but there's still a whole process to get there."
Saldich said Riverbed doesn't do local file caching because it can risk cache coherency issues when used over a large number of sites.