Remote access at warp speed

Riverbed's Steelhead Mobile WAN acceleration software client slashes Web, mail, file copy, and file transfer times for Windows and Mac users

Page 2 of 3

Fast CIFS file copies, slower by FTP
Regardless of the link speed and conditions, Steelhead Mobile matched performance with the physical appliance for most traffic. This is because Steelhead Mobile's core is based on the code found in the physical Steelhead appliance. For example, a single large file copy using CIFS over a T1 with 500 ms of latency (simulating a satellite link) took 1 hour, 44 minutes, and 16 seconds to complete without optimization. With Steelhead Mobile enabled, the same copy only took 3 minutes, 17 seconds -- a 31x improvement. Subsequent "hot" passes averaged a mere 37 seconds. Performance was the same on Windows XP and Mac OS X. Both Steelhead Mobile clients performed just as expected.

Regardless of the WAN link speed, I saw similar reductions in transfer times whether the test was a bunch of small files, a series of open and save operations in Excel, or MAPI traffic from an Exchange server. Even HTTP and HTTPS traffic benefited from Steelhead Mobile's optimization and acceleration engine. The new HTTP and HTTPS prefetch and chatter reduction proved quite effective at improving the overall Web experience.

Steelhead Mobile test results: T1 w/ 500ms RTT, 10^6 packet loss

 No optimizationCold passWarm passHot pass
CIFS -- Many small files4:08:030:26:59 0:20:280:20:23
CIFS -- One large file 1:44:16 0:02:18 0:00:310:00:28
Excel four-step45:44.60:01:190:01:000:00:51
FTP (passive)1:48:320:02:380:02:150:02:11
Note: The tests included CIFS with many small files (1,004 files but only 10.4MB in total size); CIFS with one large file (a 155MB ISO image); Excel four-step (an open-copy-save as-open process); a MAPI test that saves a 700K attachment from Exchange to the local drive; and a passive FTP test using the same 155MB ISO file as the single-file CIFS test. All tests were executed using Macro Scheduler for consistency and timing.

Not all FTP traffic is equal, and I did notice an issue when testing FTP performance. There are two types of FTP communication: Active and Passive. Active FTP is where the client initializes the communication to the server and the server connects back to the client to complete the data transfer (outbound-inbound). Passive FTP is where the client initiates the communication and establishes the communication lane to the server (outbound-outbound). Passive FTP sessions benefit from Steelhead Mobile's optimization, but Active FTP sessions do not -- even though the Mobile Controller shows the connection as optimized.

This is due to the architecture of the Steelhead Mobile client. Whereas a Steelhead appliance will optimize connections that are initiated from either side of the WAN, the Steelhead Mobile client optimizes only the connections it initiates. Mobile is smart enough not to block the FTP server's inbound connection -- Active FTP sessions do work -- but neither latency nor bandwidth requirements are reduced. Passive FTP sessions are not affected because all communication is from the client to the server. When using Passive FTP, I did see performance gains, but they're relatively modest -- an 8x improvement in the case of a single large file -- compared to CIFS file copies. Even on "hot" passes, CIFS traffic outpaced FTP response.

The Steelhead Mobile Controller gives admins an easy way to view the effectiveness of their optimization policy and the overall capacity increase on the WAN.
The Steelhead Mobile Controller gives admins an easy way to view the effectiveness of their optimization policy and the overall capacity increase on the WAN.
| 1 2 3 Page 2