Riverbed Technology Inc.s Steelhead appliances use caching, compression and protocol optimization to speed application performance over WAN links.
Click here to read the full review of Steelhead 500.
2
Riverbed Technology Inc.s Steelhead appliances use caching, compression and protocol optimization to speed application performance over WAN links.
A caching-appliance-based approach can prevent some of the performance headaches usually associated with expanding operations. IT managers often deal with these problems by throwing more bandwidth at them.
Unfortunately, given the large amounts of latency that chatty protocols such as CIFS (Common Internet File System) create, a bigger pipe doesnt necessarily help application performance.
Enter the Steelheads, which will ship this month. The appliances range in size and price from the Steelhead 500, which contains 40GB of cache space and costs $5,995, to the Steelhead 5000, which is priced at $39,995 and stores 512GB of cache. We tested only the Steelhead 500.
Like most caching devices that eWEEK Labs has seen (including Tacit Networks Inc.s I-shared cache server, among others), the Steelhead appliances cache data transmitted over the WAN into local appliances.
The Steelhead devices are useful for companies that want to centralize data for simpler management and ease of user access. With Steelhead appliances in place, data can be centralized at the home office while remote offices can access data using local devices.
In tests, we were impressed with the Steelhead 500 appliances ability to speed user access to CIFS file shares, FTP and Microsoft Corp.s Exchange, although it currently doesnt accelerate NFS (Network File System).
In contrast, Tacit Networks solution accelerates only file-serving protocols such as CIFS and NFS.
For our testing, we used two Steelhead 500s with a WAN simulator sitting between them. The client sat on one side of the WAN; the server sat on the other. Throughout the tests, we noted that with a warm data cache (one where data has been segmented and sent once over the WAN), access to remote files was comparable to LAN speeds.
Even in cases where the cache was not warm, we saw a slight increase in performance. This new speed can be attributed to the compression that the Riverbed appliances perform before sending data over a WAN.
The Steelhead appliances analyze and segment data and use proprietary algorithms to recognize and accelerate transactions. When a user needs to access a remote file, the Steelhead appliance closest to the user serves up the file locally at LAN speeds.
The Steelhead devices act as TCP proxies, and this ability lets them repack frames to make them more efficient.
With this optimization, it is possible to expedite protocols such as HTTP and FTP because frame optimization reduces the amount of trips across the WAN that are necessary for each transaction.
Because the Steelhead appliance stores data as segments, instead of just as files, a user can easily rename and alter parts of large files on a remotely accessed file without forcing the entire file to be sent over the WAN.
The algorithms built into the Steelhead appliances analyze segment requests to see which data can be served locally to a user. However, the appliances have a finite amount of cache storage, so they will not be as helpful for compute-intensive organizations such as movie production houses, biotechnology companies and research centers, which routinely create large amounts of unique data.
The Steelhead appliance easily accelerates access to remotely stored CAD files, spreadsheets and other large documents, which should make it a popular purchase in branch office environments.
The NICs in the Steelhead device have pass-through capabilities, which let data move transparently through the device in the event of a hardware failure. Even with the device turned off, data continued to flow through to the WAN in our tests.
Senior Analyst Henry Baltazar can be reached at henry_baltazar@ziffdavis.com.
Be sure to add our eWEEK.com storage news feed to your RSS newsreader or My Yahoo page: