![]() |
|
| F E A T U R E | |
|
|
|
November 29, 1999 |
||
|
|
How We Tested Software Distribution We ran three separate sites, related to each other hierarchically and connected over a WAN, with centralized control over the movement of files resident at the top site. From this central site we set up distribution jobs for moving files to servers and clients on the other two sites. We were not interested in software and/or hardware inventory, software-distribution job creation or software metering. Instead, we aspired to gauge the products' ability to create, control and track file transfers across the wide area. Each product we tested can back off from using all available bandwidth when moving files from site to site. To assess this ability, we skinned down the pipes between sites and observed not only whether the product was bandwidth-sensitive (each product is), but also how well the central-site-management software dealt with trickling software out over a long period of time without timing out or failing (the results were a ripping success for all three products). The WAN links were simulated using The Cloud version 1.1 by Shunra Software (www.shunra.com), a Microsoft Windows NT application that offers varying link speeds and impediments. We ran the link speeds at 512 Kbps with no packet or bit loss and no congestion, all the way down to 64 Kbps with as much as 1,000-millisecond latency and 30-percent packet loss, lasting 100 ms to 1,000 ms every 20 seconds.
To generate failures and to test checkpoint restarts of transfers, we yanked cables to remove target servers and clients during updates, all of which created tons of error messages.
|
|
|
|
PAGE: 1 I 2 I 3 I 4 I 5 I 6 I 7 I NEXT PAGE |
||












