Dissecting channel gateway performance

W hile the notion of linking LANs to mainframes is a decade old, recent developments are finally establishing the mainframe channel-attached gateway as a primary conduit for enterprise computing.

With potentially thousands of simultaneous sessions running across these gateways, network managers are raising legitimate concerns about how well these devices perform under such heavy loads.

There are several forces driving the resurgence of interest in channel gateway technology. First and foremost is the universal recognition of the mainframe as a critical and permanent element of the Enterprise computing environment. It is not the dying dinosaur that critics once proclaimed it to be.

The rapid and successful integration of TCP/IP into the mainframe environment is another driving force. At the edges, IP to SNA gateways _ including TN3270 and 3270-to-HTTP translators _ empower native IP clients to access mainframes via native SNA. Native implementations of the TCP/IP protocol as well as implementations of File Transfer Protocol FTP and TN3270 servers on the mainframe's operating system allow native TCP/IP to flow, end-to-end, into the heart of the mainframe environment.

Finally, where early channel gateways typically were single-protocol, such as SNA, add-ons to existing communications controllers, today's channel-attached gateways are designed as integral components of multiprotocol routers.

In mid-1997 IBM engaged The Tolly Group to conduct the most extensive independent tests ever run using channel gateway technology. While I viewed this effort more as a first step rather than the last word, the project produced some interesting findings. (Detailed test information is posted to The Tolly Group' s Web site, see address below.)

For starters, merely trying to build a controlled, dedicated test environment capable of running Enterprise-scale tests was a gargantuan task. Built at the IBM Systems Center in Gaithersburg, MD, the testbed was achored bya 5-CPU ES/9000 mainframe running MVS/ESA and capable of processing 275 million instructions per second. The high-end transaction test employed nine front-end processors.

Using a pair of Enterprise Systems Connection ESCON channels to funnel data between the mainframe and channel gateway-attached token ring LANs, The Tolly Group observed throughput levels that reached 88% of wire-speed by downloading batch data over TCP/IP to stations attached to eight LANs. Testing SNA transaction throughput in a similar configuration, rates of some 3,000 transactions per second - sustained - were recorded.

Anomalies abounded as well. For example we found SNA session instability and TCP throughput degradation at low load levels. This led the vendor of that gear to state unequivocally that extensive tuning by vendor specialists is required before channel gateways are up to the task of Enterprise-scale networking. If true, that is an important finding.

In addition to rerunning the previous benchmarks with new versions of gateways, the next round of testing should also explore critical areas such as simultaneous dual-protocol (SNA and IP) throughput tests and gateway connections to Ethernet/Fast Ethernet LANs.

Kevin Tolly is president and CEO of The Tolly Group. Reach him via e-mail at ktolly@tolly.com.

More Tolly on Technology columns