Silver Peak hits new WAN-optimization heights

You know what makes network administrators happy? Making efficient use of their equipment and eliminating "performance stinks" calls from end-users. Thus, investing in an effective WAN optimization and acceleration solution such as Silver Peak NX-5500 2.0 can put a big smile on your network admin's face.

WAN acceleration and optimization products improve WAN performance by reducing, or sometimes even eliminating, obstacles found in "chatty" applications and TCP/IP. TCP/IP is greatly affected by latency and bit errors on the link, and application protocols that require a large number of round trips (such as CIFS) can bring an otherwise fast pipe down to nearly dial-up speeds.

No longer the new kid on the WAN optimization block, Silver Peak has substantially raised the bar with its NX-5500 appliance. Targeting WAN speeds up to DS3, it delivers superior raw performance over varying WAN conditions.

Click for larger view.

The NX-5500 optimizes and accelerates not only TCP traffic but also UDP (User Datagram Protocol), a feature not common to other acceleration appliances. The NX-5500 includes Network Memory, Silver Peak's name for its caching technology, as well as a solid QoS engine. Happily, all of this can be managed via a surprisingly intuitive Web-based UI.

The NX-5500 comes standard with 2TB of local storage, 50Mbps of WAN capacity, redundant power supplies (my units had three each) and four Gigabit Ethernet interfaces (two for management, two for WAN traffic). I like that the NICs fail closed on system failure, preventing the appliance from blocking WAN traffic.

During the past couple of months, I tested a pair of NX-5500 appliances using my Shunra Virtual Enterprise 5.0 WAN simulator and a mix of HP servers and Windows XP PCs. My test results for this release blew away all previous Silver Peak performance results. For example, with the previous version of the Silver Peak box, a "cold" pass of a single, large file copied over a T1 WAN circuit with 500ms of latency using CIFS took more than 58 minutes to complete. The 2.0 release completes the same file copy in less than 11 minutes, far outpacing the unoptimized time of nearly three hours.

By comparison, the same cold test using Silver Peak rival Riverbed Steelhead 3.01 delivered a time of 1 hour, 32 minutes. Other tests showed similar improvements, with FTP showing the smallest amount of improvement. (Riverbed still wins this particular race.)

This time around, I added a couple of new trials to my series of tests. Among them, I employed Double-Take Software's backup and recovery utility to set up a branch-office-to-datacenter replication between two Windows 2003 Servers. Without any optimization, the replication over a T3 with 44ms of latency never surpassed 6Mbps. With the NX-5500 in line, a cold pass netted about 80Mbps, with subsequent warm passes exceeding 200Mbps -- all over a 45Mbps link. Silver Peak's Network Memory handled the high traffic speeds without any trouble.