MPC+: Autobahn for the S/390

A new feature in Communications Server for OS/390
boosts networking performance

IBM has recently introduced the latest technological advances in Multipath Channel (MPC), MPC+, as a new function in eNetwork Communications Server for OS/390 (CS for OS/390). MPC+ provides the most cost-effective networking attachment for S/390 by fully utilizing the capabilities of the channel and minimizing the cycles expended in the S/390. By combining the benefits of a high-performance transport technology with advanced I/O scheduling algorithms, MPC+ can significantly improve networking performance and reduce costs. Read on to learn about the evolution of channel technologies and more specifically, MPC, that led to IBM's new MPC+ solution.

The Basic Rules of the Road
Much like a roadway, a channel is the physical connection used to attach networks to an S/390 server. You can consider the channel as a highway infrastructure in which all roads lead to and from the S/390 system. A "subchannel" identifies the logical connection or road name or number (for example, Interstate 95 or Miami Boulevard). Just as large highways and local roads can share portions of the overall highway infrastructure, subchannels can share the channel and its capacity.

The channel protocols define the rules of the road, determining how the subchannel can be used by each partner subchannel. Although it is imperative that each partner obey the protocols for the subchannel, these rules can vary from one subchannel to the next. After all, different countries have their own rules of the road.

One Lane Road Ahead
Many of the early channel protocols used only a single subchannel for communications. This approach meant that at any point in time the subchannel could only be sending or receiving data. If you have ever come upon a "ONE LANE ROAD AHEAD" sign, you can immediately recognize the shortcomings of this approach in terms of traffic flow or throughput. Another limitation in this approach was that the subchannel was a single point of failure. Obviously, some major technology improvements were needed.

The early versions of MPC in the Virtual Telecommunications Access Method (VTAM®) addressed the "ONE LANE ROAD" problem by simply allowing multiple paths (lanes) or subchannels to be grouped for communications across the channel. Each one of these subchannels was dedicated to either sending or receiving data. The capacity of this MPC group was then determined by the number of channels allocated. As a result, outages on a single subchannel did not necessarily affect the connection—improving both throughput capabilities and availability. From a networking perspective, MPC functioned very much like a multilink transmission group to an S/390 server, with multiple logical connections appearing as a single physical connection.

The original MPC in VTAM Version 4 Release 1 supported only subarea SNA between two VTAM hosts. In VTAM Version 4 Release 3, MPC supported APPN® and HPR traffic via APPN Host-to-Host Channel (AHHC). Along the way, usability enhancements, such as the sharing of subchannel definitions between OS/390 and VTAM, were made. In addition, VTAM supported dynamic addition and deletion of subchannels from the group, which enabled users to add capacity or manage planned outages non-disruptively.

The design of the AHHC protocols and the peer-to-peer nature of APPN also enabled outboard devices to use MPC for S/390 server attachment. This valuable technology was (and still is) licensed to other networking vendors to facilitate more open access to the S/390. In this way, MPC became a highly scalable and available solution for S/390 attachment. If channel protocols of the past were considered a "ONE LANE ROAD," MPC could be described as a major highway. But even though MPC addressed many of the shortcomings in channel protocols, the best was yet to come.

The Evolution of eNetwork Multipath Channel
Although the earlier versions of MPC focused on the channel protocols and SNA connectivity, IBM realized that to provide truly state-of-the-art S/390 channel attachment, it had to consider MPC in the context of the overall S/390 software solution. In March 1997, IBM introduced the MPC+ solution as part of the CS for OS/390 Release 3.

In that release of Communications Server, IBM introduced the High Performance Data Transport (HPDT) family of services. The concept behind HPDT was to improve the buffering capabilities on the S/390 while reducing data movement. In environments where large amounts of data are sent or received from the S/390, this movement of data can consume a significant number of machine cycles. For example, when an application sends a 60 kilobyte message, up to 40 percent of the cycles can be spent just on moving the data.

However, HPDT makes it possible for an application to request a buffer with that same buffer being passed all the way through the protocol stacks without data movement. HPDT-MPC completes the picture by using this same buffer to transmit the data on the channel. It also employs new buffering techniques that enable the networking and protocol headers to be discontiguous from the user data. As a result, user data flows through the channel, unobstructed by headers.

In addition to using the HPDT family of services, MPC+ improves I/O scheduling interactions with OS/390. Once again, the result is higher throughput and continued reduction in machine cycles. After all, if you need to drive on the highway, you don't want to deal with a series of speed bumps that impact your transit time.

While this new technology has been used initially for SNA connectivity, MPC+ will eventually support multiple network protocols such as TCP/IP. These capabilities make MPC+ the optimal protocol for not only S/390-to-S/390 connectivity, but for other networking devices as well. In the near future, IBM plans to provide MPC+ on the IBM 2216 Nways™ Multiaccess Connector and 3746 Nways Multi-access Enclosure.

The Results Speak for Themselves
As part of the CS for OS/390 performance measurements, MPC+ was compared to the previous version of MPC in several scenarios. In each case, the configuration used three processors (logical partitions) on each of two S/390 servers—one functioning as the server and one as the client, attached via MPC.

In the first case, "3 Channels," the MPC group consisted of three channels with each channel supporting one subchannel sending data and one subchannel receiving data. The client side sent a 40-byte request to the server side, which then responded with four megabytes of data. The client established three sessions to the server. In the "1 Channel" case, a single session was used with a single channel. The results speak for themselves, as MPC+ drove the channel to its effective capacity by improving throughput over MPC by 16 to 23 percent (see Figure 1).

More importantly, there was a reduction of roughly 40 to 50 percent in machine cycles used on each processor in all cases (see Figure 2).

The combination of improved throughput and a smaller number of machine cycles results in a two times improvement in throughput per CPU (see Figure 3). In these sample configurations, the reduction in machine cycles could amount to over $350,000 (U.S.) in savings (based on CPU costs) for the same throughput.

Using this new MPC+ technology, we can estimate that sustained throughput greater than one gigabit/sec is possible on an S/390 G4 Server—clearly taking S/390 networking capabilities to new heights.

Enterprise-Class Connectivity
As an integral part of CS for OS/390, MPC+ has reset the benchmark in S/390 attachment technologies. It is far more than a change in channel protocols. Instead, it is a set of innovative technologies tightly coupled to OS/390 facilities. Although these technologies have been deployed initially for SNA connectivity, the future for MPC+ is in seamless support for multiple network protocols (including TCP/IP)—enabling enterprise-class connectivity between the S/390 and the network.

In this way, Communications Server for S/390 has definitely taken the primitive roadways of the past and evolved them into a new enterprise-class, high-speed transportation system with substantial improvements in throughput, response time, and machine cycles. Today, these throughput improvements are already enabling more efficient use of the channel capacity by MPC+. And they promise even more benefits in the near future.

For More Information

Visit http://www.networking.ibm.com/cms/cs3abt.html