NLM tool does finger-pointing
NLM tool does finger-pointing
Network monitoring program zeros in on applications that cause problems
NLM's Michael Gill says test tools help in finding network choke points.
By William Jackson
'Fifty percent of networking is finger-pointing,' according to Michael Gill, an electronics engineer in the National Library of Medicine's Communications Engineering branch.
Gill has found a tool to help him point his finger more accurately when a newly installed application does not perform up to expectations. Before rolling out an application, he baselines network performance with simulated traffic to identify choke points and help the systems staff determine whether any future slowdown comes from the pipe or elsewhere.Instant payoff
Gill began using the Pegasus network-monitoring program about six months ago before installing an automated indexing system developed at NLM. He wanted to make sure the network connection would be adequate. It was, and knowing in advance saved NLM's parent, the National Institutes of Health, the expense of installing new lines.
'We were having serious discussions about T3s' that turned out to be unnecessary, Gill said.
NLM is now working with NASA to test a series of links to connect an anatomist in Japan with large graphics files at the library in Bethesda, Md. The final circuit will extend 100,000 miles one way with two satellite hops.
The monitoring tool is 'for an operational environment, not for stress-testing a network,' he said of Pegasus from Ganymede Software Inc. of Research Triangle Park, N.C.
The Pegasus server software, which runs under Microsoft Windows NT, gathers performance data from clients installed on desktop systems or other servers. The endpoint clients work under 19 operating systems, including common versions of Windows, Unix and Linux.
The endpoint clients execute scripts to simulate application traffic and create realistic loads across a network. The Pegasus server measures real-time performance at the application layer and can provide historical trend reports. The server has 40 scripts that simulate common application types. An application scanner can create a script for other applications, or scripts can be edited and customized.
Adding applications at NLM is tricky because the library's files run as large as 20M apiece.
Pegasus provides throughput summaries via the network manager's browser, based on endpoint client reports. The clients realistically simulate traffic.
'The main product is bibliographical databases,' Gill said. The library developed a system to automate some of the identification and indexing of data from scholarly publications. During development and testing, the indexing system resided in a building with its own 100-Mbps connections.
'They were humming away in-house,' Gill said.
But the new system will go into production in a building several miles from the NIH campus. An IP over an asynchronous transfer mode connection will serve the remote site, although there is only a 10Base-T connection inward from the router.
Gill said he wondered 'how to characterize the performance between us and them.' He did it by installing a Pegasus client on a desktop system at the remote site that executed a File Transfer Protocol 'get' command every 15 minutes, 24 hours a day.
Under that load, the connection between client and server was just adequate, Gill said. That does not guarantee the new application will work perfectly when installed, but if it doesn't, he will be reasonably sure the problem is not in the network.
Pegasus is helping assess connections that will move large graphics files between Bethesda and Sapporo, Japan, for NLM's Visible Human project. The Visible Human is a series of cross-section images of male and female cadavers that were scanned and digitized. The male images amount to 15G of data, and the female images 40G'more because those sections were thinner.
NLM staff has classified and indexed many of the flat-file data sets, but some work remains to be done. A Japanese anatomist volunteered to help finish up the job on a system that can manipulate sectional graphics. Transferring the images will take a big pipe.
'These are big images, anywhere from 5M to 10M' each, Gill said.
With NASA's help, NLM has set up a test circuit to see how practical such large file transfers are. A NASA research network with a node at the library takes them as far as Star Tap, the National Science Foundation's high-performance network connection point in Chicago. From there a Canadian research network carries them to Vancouver, where they cross the Pacific to Japan.
Pegasus clients on a desktop system in Sapporo and a server in Bethesda simulated the file transfers 24 hours a day for several weeks and came up with a baseline bandwidth of 1.3 Mbps. The choke point was a T1 line in Japan.
'Things work, but they are very slow,' Gill said. 'From a user perspective, it's not worth doing.'
The library expects to open up a bigger route that will have T3 or faster links and will use an American and a Japanese satellite to carry the data. Pegasus also will test that circuit which, if all goes well, should provide a baseline 45 Mbps.