Riverbed Updates Cascade 8.4 With Application Performance Management

When WAN optimization vendor Riverbed acquired network anomaly detection vendor Mazu Networks in 2009, it was a natural fit with Riverbed optimizing applications over the WAN and Mazu providing the application detection and monitoring. With Cascade 8.4, the integration is moving along on two fronts, with Cascade providing what the company calls end-user experience monitoring and a sensor agent for the company's Riverbed Services Platform (RSP) to gather application performance statistics from Riverbed appliances. These are both moves by Riverbed close the loop between WAN optimization, application performance management and action.

Riverbed's Cascade enhancement is monitoring the end-user experience. To quantify the end-user experience, Cascade uses a metric called "connection throughput," defined as the throughput a user receives when the application is active. For example, a remote employee using an application that is fetching and saving data over the WAN is only active when the application is actually reading or writing over the network. The idle periods occur when the user is working with the data. Measuring throughput over an hour period won't provide an accurate representation of end-user experience because the overall throughput will be low. Restricting the throughput metric to when the application is reading or writing data, Riverbed claims an accurate representation of end-user experience.

"While connection throughput will provide some additional APM data, organizations will likely still need other management elements to monitor complex application environments. Looking at the desktop and user experience as well as agents on the servers will likely be needed for true end-to-end monitoring,"  said Michael Biddick, CTO of Fusion PPT and an Information Week Analytics Analyst. Riverbed is positioning their performance monitoring as an alternative to installing application monitoring agents on servers and desktops, but the company agrees that some application performance visibility—particularly client or server performance issues—is lost with a network-only monitoring solution.

What you really get is a representation of end-user throughput which may or may not relate to experience. Many network applications are transactional  and composed of relatively small amounts of data being moved back and forth. Transactions are typically bound by time—the time it takes for a request to be transmitted, processed and then received by the client application. Some applications may be bound by capacity if they are pushing large amounts of data—tens to hundreds of Mb—such as a file transfer or even e-mail. In either case, when the traffic is processed by  a WAN optimization product, the amount of data traversing the WAN should be significantly less than an non-optimized connection.

Relating user experience to throughput is a rather strange correlation that probably has more to do with the benefit of WAN optimization—reducing the amount of traffic on the WAN—than their ability to reduce application delay. Riverbed's Steelheads, like other WAN optimizers, can improve application performance by reducing round trip and pre-fetching content, but those functions are largely secondary to WAN bandwidth reduction. When we mentioned our concerns with using bandwidth as an experience measurement, representatives said they would take user feed-back into consideration.Where the connection throughput metric is useful is when it is fed into Cascade's behavioral monitoring system. Behavioral monitoring automatically builds a profile of end-user application usage over time. You can then set alerts on deviation of the established behavior profile. For example, setting an alert when a user's application usage spikes above 25 percent of the profile indicates an aberrant condition that may need looking into. For example, a new business initiative may require more capacity than expected. Cascade's behavioral monitoring can also identify bandwidth hogs based on consumption and application.

Riverbed also added Sensor-VE, a virtual system that runs on Riverbed's RSP platform and provides application performance monitoring at the remote sites without the need for additional hardware. The Sensor-VE monitors and reports on application performance through the Riverbed appliance. This allows Cascade to measure application performance for applications that are optimized. Without a monitor in the remote office, measuring application performance in the data center won't provide accurate performance measurements because the WAN optimizer in the data center transparently proxies communications locally, resulting in inflated performance measurements and it also won't account for delay over the WAN. By measuring performance in the remote office, Sensor-VE can report on application performance between the local Riverbed appliance and the client computer.

By providing application performance reports along with WAN optimization and application acceleration, Riverbed is starting to close the loop on application performance management. The last step of the loop is taking what is learned from application performance monitoring and feeding that back into the WAN optimization and application acceleration system. Today, that is a manual process. Riverbed is cagey about announcing road map items, but plans tighter integration between Cascade and Steelhead appliances in the future.