AppareNet Bags Bottlenecks

The distributed version can be deployed on several platforms, including Hewlett-Packard HP-UX, IBM AIX, Linux (2.4 kernel), Microsoft Windows NT/2000 and Sun Solaris, and is offered on preconfigured hardware of your choice -- servers from Dell, HP or Sun. The solo version deploys the Sequencer and console on a single Microsoft Windows machine that you provide.

The management console offers a GUI for creating, running and managing diagnostic tests. It uses DOS batch files for scheduling recurring tests and a Microsoft SQL Server database to store test results (the company says support for Oracle databases is expected by year's end). The management interface lets you search and view completed tests based on various criteria, including user ID, target host name and title. Searches also can include wild cards to broaden the results.

The solo version of AppareNet costs about $30,000. The more complex distributed system starts at $75,000, and varies based on the hardware platform and number of Sequencers you purchase. AppareNet's value lies in its ability to detect network misconfigurations that may point to the cause of network congestion or poor application performance, possibly obviating an infrastructure or bandwidth upgrade, or assisting in forcing a service provider to address performance issues.Gone Fishing

A beta version of AppareNet 1.5 arrived at our Real-World Labs® in Green Bay, Wis., pre-installed on a HP Compaq 1U server and equipped with both a 10/100 3Com NIC and an SMC GigEthernet NIC.

I connected the AppareNet server to the network and booted the machine. Using the 10/100 connection, I configured a test by selecting a Sequencer and a target -- one of the machines on the local segment. Test parameters included the number of iterations, sleep time between bursts and the testing mode. You can choose between testing only the target or testing all hops between the Sequencer and the target.

The system builds a statistical model of the network based on the modification of packets as they traverse the network to and from the Sequencer: statistics on packets sent and lost, bandwidth (two-way in kilobits per second), percent of pipe utilization, propagation delay (one-way in milliseconds) and jitter rates (one-way in milliseconds).

I started the tests and checked out the results and updates displayed on the management console as they were received by the Sequencer.

The 100-Mbps link was achieving only 82-Mbps on the local segment. When the test finished, the condition column of the test results indicated a problem. A quick click on the red "X" displayed a detailed diagnostic indicating a probable duplex mismatch. I was given the choice of making a change or letting ApparaNet autonegotiate the problem. I forced the port that AppareNet was connected to on the Cisco switch to full duplex, 100 Mbps, and reran the test. The results indicated total throughput of 195 Mbps, the maximum achievable rate for a full-duplex 100-Mbps link.

Good Catch

Additional testing with the AppareNet on the Internet provided fascinating results and pointed out issues with backbone providers and individual hosts. Scheduled tests can be used to verify QoS (Quality of Service) and bandwidth guarantees from your service or hosting provider.

AppareNet's ease of use and deployment belies the power within the system. It is worth a look if you're interested in uncovering the inefficiencies in your own network or even mine.

Technology editor Lori MacVittie has been a software developer and a network administrator. Most recently, she was a member of the technical architecture team for a global transportation and logistics organization. Write to her at [email protected].