Packet loss slows thin-client file transfers

Tests reveal correlation between increase in file transfer durations and packet-loss increases over Citrix ICA client sessions.

By Charles Bruno
Network World, 09/18/00

The duration of file transfers flowing over remote thin-client sessions potentially can double or even triple when packet-loss conditions rise even slightly, according to recent tests conducted by Tolly Research.

The findings reveal a direct relationship between packet loss and file-transfer performance over remote thin-client connections. The impact, however, was not nearly as great as the impact of latency on file transfer seen in earlier tests conducted by Tolly Research.

Engineers examined bandwidth consumption and duration of execution for a 1M-byte file transfer over a simulated T-1 connection with configured packet-loss conditions. The tests were conducted in a Windows NT environment using Citrix MetaFrame 1.8 as the application server. For the client station, researchers used the Win32 and ActiveX ICA clients only.

Citrix's Java client does not offer native support for file transfers. Packet-loss percentages used for these tests were representative of boundaries set for performance on domestic backbone networks. Reported packet loss is typically less than 1%. Tolly Research used values of 1%, 3%, 5%, and 10% for packet loss. For the purpose of testing, packet loss was configured as a fixed percentage.

Both clients (Win32 and ActiveX) displayed a steady increase in transfer time and amount of data transferred (decrease in bandwidth consumption) that corresponded to a shift in packet-loss level.

The clients' overall performances were very similar. The ActiveX client always required more total bits to complete the transfer at each level. This was a consequence of much lower average packet sizes. The execution time was fairly close until the 10% packet-loss level, where the ActiveX client slowed down considerably.

Relative to performance on a loss-free network, file-transfer times essentially doubled when packet loss reached a level of about 3%. Depending on the ICA client, file-transfer durations doubled or tripled as packet loss extended from 3% to 10%.

While these results are significant, the outcome does not warrant the same level of concern that network latency demands for file transfers.

"Consider that the degradation in file-transfer performance observed for a packet-loss level of 10% is roughly equivalent to that experienced on a network with a 50-ms to 100-ms round-trip latency," says Kevin Flood, chief technology officer of Tolly Research. "And a packet-loss level of 10% is almost never seen on an Internet backbone while round-trip latencies of 50 ms to 100 ms are quite common."

The above information is derived from an ITclarity research track report. To purchase a subscription to Tolly Research's ITclarity service, click here.

About Tolly Research

Tolly Research, provider of ITclarity, is a technology research firm based in Brielle, N.J., that specializes in testing-based research. Tolly Research provides accurate test results, insightful analysis, and practical recommendations for leading-edge technologies required by network professionals. While many businesses have specific implementation issues, there is a broad base of technology research that is applicable to all organizations. Tolly Research offers a more cost-effective method of gathering data than committing internal resources to tackle each new technology. For more information on Tolly Research, point your browser to www.tollyresearch.com, send e-mail to itclarity@tollyresearch.com or call 732-292-9411.

More report summaries from Tolly Research:

Proprietary VoIP encoding falls short

Low-bit-rate voice encoder gets VoIP nod

Link aggregation tests end in server crash

Tests yield impressive voice quality

Citrix VideoFrame excels

Citrix ICA survives packet loss

Researchers find fault with Win2K TCP/IP gateway services

Tests show Microsoft's Terminal Server, RDP hog bandwidth