Special Delivery: Fast WANs and App Acceleration

Technology executives are centralizing data while their employees are dispersing. They're all waiting for the WAN to quickly deliver hot applications to mobile workers anxious for performance, and vendors are racing to deliver.

Technology executives are centralizing data while their employees are dispersing far and wide. Theyre all waiting anxiously for the WAN to deliver hot applications—fast—to mobile workers hankering for performance.

Technology managers and vendors are racing to oblige. This perfect storm of needs—including more mobile workers, the Webification of software, speedier networks and office consolidation—is driving the rapid growth of application acceleration technologies to the point where they are becoming enterprise mainstays just as switches and routers are today.

As those technologies, delivered today mostly as network appliances, become strategic for enterprises, Juniper Networks is one vendor capitalizing on the trend. On Oct. 16, Juniper will introduce a new high-end appliance and operating system update aimed at enterprises looking to increase their deployments from a dozen to hundreds of locations while ensuring high availability of the devices.

Junipers launch of the new WXC590, which can support 140 locations over a single 45M bps DS-3 line or 420 locations over a 155M bps OC-3 connection when three are stacked together, coincides with the launch of another new high-end appliance from upstart rival Silver Peak Systems.

Silver Peak will introduce its NX8500 appliance, capable of supporting up to 500M-bps data rates for replicating data between data centers in business continuity applications (see related story, Page 24).

Technology executives and analysts say whats driving the rapid acceptance of these technologies—which are divided between ADCs (application delivery controllers) that sit in front of Web servers and WOCs (WAN optimization controllers) that work in pairs between a central site and each remote branch office—can be attributed to many factors, but one sticks out: Information workers typically operate away from the corporate LAN.

Companies are removing equipment such as file and print servers from remote branch offices and centralizing them in a data center to reduce costs and improve manageability and security. This also helps to ensure that backups actually happen—a necessity for compliance with regulations such as the Sarbanes-Oxley Act.

The problem: Most applications are poorly designed for operating over lower-bandwidth, higher-latency WAN links. In addition, backing up large volumes of data over WANs running TCP/IP is not practical without acceleration techniques.

Increasing the capacity of WAN links—throwing bandwidth at the problem—doesnt help, since the issue is latency, which can increase with distance.

Executives at one large enterprise said the company began evaluating acceleration technologies more than three years ago, when sensitive customer and employee data theft issues began dominating the headlines.

"It was perfect timing that this technology we had been playing with was right at the forefront of allowing us to do backups over the network without having to buy a very expensive and recurring [charge] WAN upgrade," said a large Riverbed Technology customer, who asked not to be identified.

"The problem we solved was [the need for] more tape backup. We can do backup to a central data warehouse over the network without buying more [WAN] capacity." The customer plans to deploy hundreds of Riverbeds Steelhead WAN optimization appliances.

That enterprise found out the hard way that throwing bandwidth at the problem doesnt work. A Steelhead appliance in one location was moved elsewhere, and the company replaced its 1.544M-bps T-1 link with a 44M bps T-3 link.

"Its a 28-times-bigger pipe, but the users complained it was slower [without the Steelhead appliance]. It was a latency problem, not a bandwidth problem," said the user.

Chatty Protocols

In moving file and print servers out of remote branch offices, other enterprises are encountering latency issues created by the chattiness of legacy protocols such as Microsofts CIFS (Common Internet File System) and MAPI (Messaging API).

Applications such as Microsofts Word, PowerPoint, Excel and Exchange using CIFS and MAPI break up the response to client requests for file access into smaller chunks that are sent sequentially.

The client acknowledges each received chunk, and the next one is not sent until that acknowledgment is received by the server. So a simple action such as dragging a file from a remote file share to a local desktop generates as many as 3,000 or 4,000 client/server interactions.

Analysts at NetForecast conducted a study earlier this year of application performance response times across varying distances for CIFS, MAPI, Web and SNA applications. The study found that CIFS and MAPI performance "drops precipitously with even small distance increases from the server," author Peter Sevcik at NetForecast, in Charlottesville, Va., said in the report.

"With servers centralized in New York, performance for both application types would be poor for users in Chicago and utterly unacceptable for users on the West Coast. Without ameliorating steps, stretching user-to-server distance for CIFS- and/or MAP-based applications plots a sure course to an application performance shipwreck."

/zimages/6/28571.gifCisco ups the 10 Gigabit ante. Click here to read more.

But the problem is not limited to Microsoft applications. Web applications in the study fared well at distances of up to 2,000 miles, but then response times dropped precipitously, and by 3,000 miles were considered to be in the poor range.

Web-based HTTP applications also suffer from chatty protocols and redundant transmissions.

"A browser will do individual fetches to get all the icons on your screen. If there are big objects, a fetch might take two or three round trips. If you refresh the screen and only one thing has changed, you still download the whole screen," said Joe Skorupa, an analyst with Gartner, in Stamford, Conn.

WOCs solve those problems and boost application response time over WAN links using a variety of techniques. The most common grouping of techniques includes sequence caching, compression, protocol spoofing, TCP optimization, QOS (quality of service) enforcement and encrypted tunnels.

Their counterparts in the data center—application delivery controllers—address more server-based performance issues by offloading from the server functions such as load balancing, TCP connection management, SSL (Secure Sockets Layer) encryption and Web application firewalls. They also execute compression and caching.

Next Page: Consolidation crush.