April 16, 2008 | Comments: (0)
Why a change to an AutoCAD file format is throwing some WAN accelerators for a loop
An innocent change to an AutoCAD file format has cut sharply into the ability of some WAN acceleration solutions to speed the transfer of these files, and caused grumblings in some widely distributed AutoCAD shops over slowed WAN performance. The problem affects users of AutoCAD 2007 and 2008 (specifically those who open and save the design program’s files over a WAN) and stems from a change to the DWG file format for AutoCAD 2007.
Ironically, Autodesk’s changes to the DWG file format were designed to improve performance and reduce file size, among other things. But the new format, when combined with the automatic backup setting called Incremental Save Percentage (ISP), has quietly turned the WAN optimization and acceleration industry on its ear, or nearly so. See the recent bulletin from Autodesk and Riverbed Technology in the Autodesk Knowledge Base.
How does a DWG file negate the benefits of WAN acceleration? Most WAN accelerators reduce data on the WAN through a process called data deduplication. As data passes over the WAN and through a pair of WAN optimization appliances, byte segments are stored and hashes or tokens are created to represent the recently seen data. On subsequent passes, instead of sending every byte of a file back over the WAN, only the tokens are sent that match data already in the local data store. This dramatically reduces the amount of data traveling the WAN while improving response time and performance for the user. Many WAN appliance vendors use this or similar techniques to great advantage.
The DWG problem boils down to this: AutoCAD 2007/2008 files are completely rewritten when a user does a full save on the drawing with AutoCAD’s Incremental Save Percentage value set to 0. Instead of only a few bytes of the file changing with a simple edit, the whole byte structure of the file changes. So to appliances that rely heavily on deduplication and pattern matching, each file save looks like a cold pass, gaining little data or time reduction from the WAN accelerator. It is important to note that this issue only impacts the save operation. Subsequent reads are not affected and will still benefit from data deduplication and overall better performance.
AutoCAD users can mitigate the scrambling effects, to a certain point, by changing the Incremental Save Percentage to 50 or higher (100 is the max). In fact, AutoCAD ships with the setting at 50 out of the box. In years past, AutoCAD users experienced some file corruption with any ISP setting greater than 0, so for many, changing that value to anything else is out of the question. But there will be times, even with an Incremental Save Percentage of 100, that each byte will be changed during a file save operation.
As a result of this file rewriting, a bit of a cat fight has broken out between leading vendors Riverbed Technology and Silver Peak Systems. Riverbed is affected by the AutoCAD file format/ISP setting issue more so than rival Silver Peak, and Silver Peak would like the world to know. To be clear, Riverbed will still accelerate AutoCAD DWG files over the WAN, even on a full save, but the data reduction is less than what Silver Peak is able to accomplish.
The AutoCAD file problem is a very specific use case and not an indicator of any larger technology issue with Riverbed. Silver Peak’s data deduplication technology is able to handle this particular problem much more gracefully than Riverbed’s.
I recently participated in an online demonstration with Silver Peak where I saw the effects of the rewritten AutoCAD files when saved over the WAN, and how to a large extent, Silver Peak’s appliances still provided a good measure of performance increase. The following table summarizes the test results. The lower the ISP setting, the more the effects of the rewrite are felt. The tests were conducted over a 10Mbps network with 100ms latency. Times are in seconds. The DWG file was 14MB in size.
| Incremental Save Percentage | 0% | 50% | 100% |
| Base line | 40 | 65 | 62 |
| Optimized save (cold pass) | 39 | 62 | 61 |
| Optimized save (warm pass) | 23 | 20 | 19 |
| Optimized save (hot pass) | 23 | 20 | 18 |
| Improvement | 43% | 69% | 70% |
Source: Silver Peak Systems
I was not able to run the same tests with a pair of Steelheads from Riverbed, but they claim to provide at least a 20% improvement in performance even on a “cold” file save.
One thing to remember is that other factors in addition to data deduplication are at work to increase WAN performance. Mitigating the effects of latency, reducing application chattiness, and overall TCP optimizations all play a part in speeding up file transfers over the WAN. Riverbed executives stated they are working closely with Autodesk to minimize or even eliminate this problem, but no time table on a fix was available.
Posted by Keith Schultz on April 16, 2008 10:58 AM
April 07, 2008 | Comments: (0)
Test Center Tracker: Packeteer sizzles at CIFS; RIA development heats up
WAN speed record: For several years running, our testing of WAN acceleration appliances has served mainly to chronicle the superiority of the Riverbed Steelhead, whose approach to byte- or segment-level caching and CIFS optimization has made it the perennial performance leader and our annual Technology of the Year Award winner. Only Silver Peak Systems, which inched closer year by year, could give Riverbed a run for its money. Last week we discovered that speedy wide area networking, or at least the branch of WAN acceleration concerned with file transfers, is a three horse race. Packeteer's iShaper may not be a better overall solution than Riverbed just yet, but it registered the best CIFS performance in our testing to date. See Keith Schultz's review.
Adobe AIR is the answer? Adobe AIR is not yet widely known or implemented, but it solves all of the major issues keeping the browser from being a common front end for applications, says Tom Yager. Read AIR's praises in Tom's "Ahead of the Curve," then weigh Martin Heller's counterpoint in "Strategic Developer."
RIA for the enterprise: Curl's longtime focus on creating rich, Web-based business applications has paid such dividends as excellent performance, smooth handling of intermittent connections, and support for large data sets. Version 6.0 of the InfoWorld Technology of the Year Award winner advances with skinnable controls, more sophisticated graphics rendering, a Macintosh runtime, and the ability to add a Curl applet to an AJAX page, and vice versa. See Martin Heller's review.
Beat the heat: "In many cases, a datacenter can generate enough heat to heat a building 10 to 30 times its size," says says Steve Sams, vice president of IBM Global Site and Facilities Services. Instead of casting that heat to the winds, some companies are using it to keep other buildings warm, and even to generate electrical power. See Ted Samson's "Sustainable IT."
It's the applications, stupid: Thanks to today's more secure operating systems, remote attacks, where the end-user is not involved at all, are becoming almost a rarity. That means educating end users is the key to client security. See Roger Grimes's "Security Advisor."
Smaller disks, lower power: Mario Apicella tackles a new twist on the old speed versus capacity question: the advantages of arrays built around larger capacity 3.5-inch drives versus those that leverage newfangled, lower-power 2.5-inch disks. See "Storage Advisor."
Posted by Doug Dineley on April 7, 2008 09:14 AM
October 20, 2006 | Comments: (0)
InfoWorld has not published a comparative review of WAN optimization solutions, but it is not because we didn't try. Back in early 2005, just before Juniper's agreement to acquire Peribit kicked off a buying spree in the space, we made plans to round up a bunch of WAN products, expose them to a variety of traffic types and link characteristics, and see how things shaked out. We extended invitations to many vendors, and we discussed test methodology with a few, but in the end only one or maybe two were willing to be measured against competitors. Nearly all were eager to support a stand-alone review. But a comparison? The answer was either "Nope" or "We'll get back to you."
Some of the vendors offered what you might call artful excuses. Here's a list of the jucier ones we heard:
Because most WAN links have no appreciable packet loss, our solution doesn't handle that and we'd prefer you not test us that way
We don't want to be compared to vendor XYZ's CIFS optimization, although we are the best at CIFS optimization
Our focus is on optimizing long, fat pipes, not short skinny pipes, or short fat pipes, or long skinny pipes
We don't want to confuse our customer base by being compared to vendor XYZ
We do WAN optimization, not WAN acceleration (or vice-versa)
Posted by Doug Dineley on October 20, 2006 01:13 PM
September 13, 2006 | Comments: (0)
Riverbed 3.0 stirs more spice into WAN soup
When I first looked at Riverbed's Steelhead WAN appliance, I found it an impressively easy-to-set-up, streamlined WAN accelerator.
When I reviewed version 2.1.2 a year later, the results proved that my previous performance stats were no fluke, and the addition of Proxy File Service allowed specific performance enhancements for MS SQL and continuous service even if the WAN is down.
Now, after getting the chance to examine the third version of Riverbed's appliance, I'd have to say it's getting even better. The already-impressive performance is supplemented with another speed boost and a wider range of features.
Riverbed recently announced six new WAN acceleration appliances, five of which scale from 15,000 to 1,000,000 optimized TCP connections and up to 4Gbps of optimized throughput, far exceeding the 45Mbps of the last release.
New to RiOS 3 (Riverbed's operating system) are additional CIFS, MAPI, and NFS optimizations, and QoS for all TCP and UDP traffic. One of the biggest gains is in NFS performance. Riverbed's NFS application streamlining improves NFS performance by reducing its chattiness, allowing for a nearly ten-fold improvement over RiOS 2.1, the company says.
The new release includes policy-based QoS for all traffic going out over the WAN, both optimized and pass-through. Per-interface and per-class policies are now built-in. Also new is the ability to prioritize traffic based on bandwidth and latency, providing greater protection to real-time and interactive traffic. RiOS 3 will also address asymmetric routing issues on the client side, helping to simplify integration for remote sites with dual WAN routers and Steelheads.
Not to be ignored, RiOS 3 provides more visibility into WAN performance by exporting to NetFlow and comes with more flexible reporting. The reporting engine in previous versions was a bit anemic, but release 3 looks to give better feedback to network admins. I will have hardware in hand shortly to test and will see just how fast RiOS 3 can go.
Posted by Keith Schultz on September 13, 2006 09:59 AM