Exclusive: WAN optimization, the Cisco WAE

As the enterprise becomes more dispersed and applications continue to greedily gobble up precious WAN bandwidth, IT struggles to find ways to coax just a little more performance out of its existing circuits. All of the bandwidth in the world won’t reduce application response times as long as TCP and latency remain tightly bound together. In order to improve WAN performance, it takes digital sleight of hand to work a little WAN magic.

Enterprises that “buy only Cisco” infrastructure now have a solution to curing WAN performance woes caused by latency, inefficient application protocols, and oversubscribed circuits: Cisco’s WAE (Wide Area Application Engine). Using a combination of TCP optimization, proxy services, and byte-level and file caching, WAE does a good job speeding up sluggish networks and significantly reducing response times, even on first-time passes. Graphical reporting and monitoring provide adequate insight into traffic patterns and WAE effectiveness.

Back in rack

I received a full rack of Cisco equipment during my evaluation of the WAE — basically a distributed enterprise in a box. I tested the WAE 512 1U appliance (1GB RAM, 250GB disk; expandable to two disks) with a standard load of Cisco’s WAAS (Wide Area Application Services). Also available are the WAE 612 — 2GB RAM, as many as two 300GB disks, and the WAE 7326 — 4GB RAM, dual processor, and as many as six 300GB disks, each one providing greater scalability. The WAE 512 is intended for small- or branch-office deployments; the other two scale to the datacenter.

With the current release, the WAE connects out-of-band with the network, and it requires either policy-based routing or WCCP (Web Cache Coordination Protocol) v2 — the preferred method — to send TCP traffic to the appliance. WAE does not optimize UDP (User Datagram Protocol) and will pass it through unchanged.

The next version due out in early 2007 will allow for transparent inline installation. Furthermore, it will be available as a network module for ISR-ready Cisco hardware, making it easier for admins to get on the WAN optimization bandwagon with a simple installation into existing hardware.

To see how well the WAE performed, I integrated my Shunra VE and Windows SBS (Small Business Server) 2003 Server into the Cisco mix to allow me to generate test results consistent with previous reviews. After the bits had settled, I found the Cisco solution to be right in the mix of all of the leading WAN optimization appliances but lagging ever so slightly behind Riverbed’s Steelhead solution.

WAE’s CIFS performance was at its best when dealing with many small files, but only average when transferring a single large file, regardless of link speed and latency. One of my tests gauges CIFS performance during multiple file opens, edits, and saves. WAE showed marked improvement compared with nonoptimized connections (sometimes as much as 30 times faster), but only average compared with other WAN vendors’ solutions. MAPI performance showed good improvement compared with nonoptimized traffic, whereas FTP traffic — a single large ISO file — did better than nonoptimized but lagged results posted in my recent tests of Riverbed’s Steelhead.

Chunky bits