Web/E-business /

Law firm defends move from software to hardware caching

Better performance drives Perkins Coie from software proxy servers to CacheFlow's Internet Caching Appliances

By Suzanne Gaspar
Network World, 04/10/00

The evidence was persuasive. After a trial that lasted several weeks, the law firm of Perkins Coie issued its verdict: It would switch from software-based proxy caches to hardware-based appliances from CacheFlow.

The Seattle based firm made the change because it wanted to increase its cache hit ratio, cut down on WAN bandwidth, provide better service to end users and make remote administration easier. Perkins Coie also wanted to postpone having to add another T-1 line.

E-mail, database access and billing are the bread-and-butter WAN applications for Perkins Coie, says Nate Lynch, senior technical analyst for the law firm, which has a 15-site frame relay network. When Lynch replaced the Microsoft proxy server software in its central office with CacheFlow's Internet Caching Appliance 1000 last March, the results were immediate. Cache performance improved, while Web traffic decreased, freeing up more bandwidth for e-mail traffic and local users.

In fact, the ratio of Web traffic to e-mail traffic flipped from 60% Web and 40% e-mail to exactly the reverse. And the cache hit ratio is now at 65% to 70%, a step up from the 50% previously achieved with the software proxy server. This hit ratio increase means more Web page requests are being served from the cache as opposed to being served from the expensive WAN bandwidth of the Internet.

Lynch says these benefits are attributed to the CacheFlow 1000 being a dedicated device with an operating system that was designed specifically for caching.

The research, planning and budgeting to put the CacheFlow 1000 in the central office took two months. Once purchased, the CacheFlow 1000 was plugged in and working in under two hours. IT is still in the process of procuring and rolling out smaller CacheFlow 110's to the fourteen remote offices; one is currently installed and three are in progress. Each CacheFlow 110 needs to be configured to work with the CacheFlow 1000 in addition to the remote site's specific network speed. Once configured, it's basic plug-and-play for the remote site support staff.

The play part is simple. Remote office browsers are configured to a CacheFlow 110 proxy. When users request a Web page, the CacheFlow 110 checks its cache and delivers the page if it's available. If not, the request goes over to the CacheFlow 1000, and if the page is in cache, it gets immediately sent to the user. Otherwise, the request will go out from the CacheFlow 1000 to the Internet, the page gets cached on the 1000 and then sent along to the user.

Lynch says the firm's existing Microsoft proxy servers and CacheFlow appliances work well together. However, the Windows NT servers running the Microsoft proxy need upgrading, and replacing them at $5,000 to $6,000 a pop is costly. The base price for CacheFlow 110 is currently $4,495. In comparison, CacheFlow 110's cost is defendable because remotely administering the NT servers that run Microsoft proxy can be problematic; a whole lot of things can go wrong. Conversely, CacheFlow's appliances offer a Web-based graphical user interface allowing IT to remotely reboot, view utilization reports and configure refresh factors.

Lynch says the CacheFlow's product has proven to be strong and reliable, with no failures reported to date. In the event of a power failure, the cache appliance can be back on within two minutes. The CacheFlow operating system writes across all disks, so if all but one disk fails, the CacheFlow still functions.

Perkins Coie spent a lot of time doing researching on the Web before selecting CacheFlow, which stood out because of its performance features. In fact, Perkins Coie downloaded a perl script from CacheFlow's Web site and used it to measure the performance of its software proxy cache.

The cache performance-testing tool is a command-line program used as an HTTP load-generating client and as an end-user response time measurement client. It records several kinds of data into a log file that are used for cache performance analysis, the response time for and the size of each object and complete page retrieved. It periodically reports the average rate at which objects are being retrieved from a cache and data is used to measure the end-user page response time of a cache under a variety of background loads.

After using this tool to check the performance measures of their Microsoft proxy server, IT realized they needed stronger caching performance and called in CacheFlow to do a demo. With the CacheFlow demo onsite for two weeks, IT used the tool to test the CacheFlow 1000 performance measures and the evidence sold them on CacheFlow.

Lynch's only advice is to buy more disk and RAM cache than you think you need because it gets used up quickly. Within a month IT was amazed they were able to fill up 40G bytes of disk cache. Having more disks and RAM makes things more efficient, because cache RAM can serve pages quicker than any hard drive. The more disk cache available, the more pages can be stored locally and pulled from cache, thereby increasing cache hit ratio.