The Citrix NetScaler handles standard load-balancing functions, but is billed as an Internet application accelerator, due to the features that speed up delivery of browser-based applications, including caching, TCP buffering, TCP session consolidation, compression, and SSL offload. These technologies can have a tremendous impact on Web application performance and reduce network traffic as well. The NetScaler is an extremely capable load balancer, an application-layer firewall, and an application accelerator. In fact, it has more features and functionality than can be covered in a review. The administration and configuration guides fill two volumes and more than 1,000 pages.
Although it is tempting to try to come up with a hard number to characterize performance gains from these technologies, the fact is, your mileage will vary. Each of the components can have an impact varying from small to huge, depending on the type of Web application, the type of data being moved from server to client, the network and server load conditions, and more. Most users will see an improvement of at least 400 percent (4X, or a reduction in time to display a page to one quarter the original time). As much as 200-fold improvements are possible, especially if the Web servers and network are heavily loaded.
Click for larger view. |
To run the NetScaler through its paces, I set up a virtual cluster of several servers to serve a demo version of an e-commerce site. The servers varied in processor number and power. I then used an Ixia 400T traffic generator and IxLoad software to simulate a large number of users accessing the Web site and compared the loads generated on each Web site. The NetScaler 10010 appliance was able to keep actual loads on the servers consistent even though their processing power varied considerably. I then enabled a number of features, such as SSL sessions, acceleration, and application security, and attempted to overload the load balancer by simulating many simultaneous users. This was possible only with artificially small sessions; when simulating actual traffic, the gigabit connection became saturated before the limits of the device were reached.
Testing the Web application acceleration features is more problematic. The greatest improvements in delivery of pages from the Web servers will occur when the server is heavily loaded (or would be heavily loaded if the NetScaler weren’t present). The type of content being delivered also makes a big difference. Static HTML pages will see big improvements but dynamic pages using .ASP, CGI, Java, or other application servers may not seem much faster, because the performance of the server delivering the application, and not network performance, will be the gating factor.
All that said, I tested page response times under a variety of conditions and found improvements from 1.5- to 200-fold. Using a mix of HTML, graphics, and active pages, I found an average improvement in the response at the client exceeding 5X; specifically, the average response at the client went from 1.2 seconds to 0.193 seconds. This is a perceptible gain and would be greater on heavily trafficked sites.
Acceleration ABCs
The NetScaler supports both integrated caching and the use of an external cache server. In either case, requests for content
coming from clients are served from the cache rather than from the Web server. A multitude of options ensures that frequently
changing content is fetched from the Web server rather than cache, or that the cached data is updated regularly. You can choose
to cache only static content or use a flexible set of rules to determine when to cache dynamic content.
TCP buffering is useful because the network between Web servers and the NetScaler is invariably much faster and more responsive than the Internet. By holding responses from Web servers until clients are ready to receive them, the NetScaler reduces the chattiness between Web servers and clients, freeing server memory and lowering CPU utilization. If the Web server is heavily loaded, this can make a dramatic difference; on lightly loaded servers, the effect is not noticeable.
Logan G. Harbaugh is senior contributing editor of the InfoWorld Test Center.
Talkback
E-mail
Printer Friendly
Reprints



