Array of Light

The Array TM is a modular system, which means you can purchase only the functionality you require. The base product provides server load-balancing, firewall functionality (WebWall) and clustering. For $6,000 per module, you can also purchase caching, compression and GSLB (global server load-balancing). An additional $8,000 will get you SSL acceleration--not a great deal but nice if you want to manage all your traffic from one spot.

A redundant configuration for high-availability environments is also available, and though an active-active configuration is possible, the KISS principle employed by Array Networks disallows an active-active configuration for the same VIP (Virtual IP address)--and stateful failover is not an option. If you only want to balance a single VIP, the redundant configuration means active-standby mode only.

CLI (command-line interface) jocks will feel right at home with the Cisco IOS-like interface. The device is a breeze to configure and operates in a standard reverse-proxy configuration. Using TCP multiplexing to back-end Web servers enhances the product's performance. SSL re-encryption to back-end secure servers is also available and lets you take advantage of pooling HTTPS (HTTP Secure) sessions on the back end to reduce the latency introduced by establishing secure connections.

Layer 7 functionality is limited compared with that of the competition, including F5 Networks and NetScaler. Only specific HTTP headers are available for routing: host, URL, network and cookie. Plans to support additional headers, such as the SOAPAction header used by Web services, are in the works. URL parsing is rudimentary, offering limited regular expression pattern-matching. When questioned about the limits on content switching, Array Networks' engineers explained that only by limiting the possibilities could they properly optimize their code to reduce performance degradation caused by excessive data parsing.

The Flight Deck, Array Networks' GUI dashboard, is a boon. At a glance, administrators can ascertain the performance both of the device's internal systems and its load-balancing functionality. The rest of the interface design is slick and navigable, but there are usability issues--too many "apply" buttons on a single screen with no apparent grouping of controls makes configuration via the GUI confusing. Array Networks says it is addressing this problem, and though the GUI has been improved dramatically since earlier releases (see our Sneak Preview of the Array 1000), I strongly advise administrators to stick to the CLI except for monitoring. SNMP-, syslog- and CLI-based monitoring options are also available, sans sexy graphics.Although I experienced no problems during testing, the proprietary OS has a unique approach to preventing DoS (denial of service) attacks in its TCP/IP handshaking process that can reject valid connections.

Does It Perform?

I inserted a single Array TM device configured for caching into our network and configured a virtual service served by a Caw Networks' WebReflector. The Web-Reflector was set up to provide three virtual Web servers, each serving up a 1-KB file. Load generation was provided by Caw's WebAvalanche, with each virtual user performing a single transaction consisting of 15 requests. This was designed to simulate a single page request comprised of 15 different 1-KB Web objects. I turned on the Array TM's caching feature and barraged it with requests.

We reached the limits of the Avalanche when it reached just over 27,000 HTTP 1.1 gets/second. While the Avalanche test was running, I conducted additional tests with Microsoft's Web Application Stress Tool (WAST) and ApacheBench. The Array successfully performed an additional 1,500 gets/second while maxing the capabilities of the two additional client machines. Dropping the number of Web objects in a transaction from 15 1-KB objects to one 14-KB object caused performance to decrease to 8,000 gets/second and reduced CPU utilization on the device from 80 percent to 30 percent. Turning the cache off resulted in 6,000 gets/second to retrieve a single 14-KB file.

I enabled the compression feature on the same 14-KB file and was delighted by the result. Rather than transferring 14 KB per request, the device transferred a mere 106 bytes, dramatically reducing the amount of traffic on the wire. Your results will vary, of course, depending on the content being served. The Array TM uses RFC standard gzip compression based on the client's ability to accept such content.The Array TM is definitely a step in the right direction for Array Networks. It performs well, its base price is competitive and the option to add modules such as caching and compression inexpensively more than makes up for the product's less-than-robust Layer 7 routing capabilities. Technology editor Lori MacVittie has been a software developer and a network administrator. Write to her at [email protected].