In-Depth

SNA over IP: The Best of Both Worlds

• 08/01/1999

Most of today’s business data resides in mainframes. In fact, in many businesses, almost every online transaction runs through a mainframe at some point, mostly through SNA networks. Customers are looking hard for ways to access this wealth of data and applications as they consolidate their network protocols, typically converging on TCP/IP.

What’s more, accessing SNA applications over IP has become increasingly critical as enterprises have attempted to integrate their host applications with browser-based "Host Publisher" or pure browser-based host access. Whether Internet, intranet or extranet focused, the terabytes of data found on mainframes and accessed through SNA applications demand a workable means of combining SNA and IP.

For many medium-sized and large businesses, however, SNA and IP remain locked in separate worlds. A common strategy has been to segregate and minimize traditional SNA applications, while only investing new resources in TCP/IP-based applications and hardware, such as TN3270 servers and IP routers. This "either SNA or IP" mind-set has been expressed by maintaining either separate TCP/IP and SNA networks, by running "dual" SNA and TCP/IP networks as sub-areas over a common backbone or by investing heavily in TN3270 to eliminate SNA from the clients/WAN.

Any dual network configuration adds cost to the enterprise by requiring additional routers, system programmers and software. Rewriting SNA applications is hard to cost-justify. And it may be impossible to find the original applications developers or equivalent skill sets to rewrite the applications.

While many technologies available today help eliminate parallel networks, the most important ones are Data Link Switching (DLSw) and Enterprise Extender (EE). DLSw has been around for several years. EE is relatively new. While both succeed in delivering SNA applications over IP, they approach "SNA over IP" in fundamentally different ways.

DLSw is a router-based solution using an extended form of the LAN bridge function that interconnects local area networks across wide area networks (WANs). When two SNA nodes on separate LANs are interconnected through WAN connections (through two LAN bridges), session time-outs frequently occur because of buffer and bandwidth shortages in the LAN bridges and on WAN connections. To prevent time-outs, the logical links between the two SNA nodes are divided into three segments: one segment in the middle between the two bridges (functioning as DLSw nodes) that are interconnected through TCP connections, and two segments between the SNA and DLSw nodes on each LAN segment. DLSw nodes pace traffic separately on each segment, acknowledge immediately when they receive a frame on a segment and move onto the next segment without waiting for transmission to be completed.

Most of the leading router vendors, including IBM, Bay and CISCO, support DLSw. They will provide performance benchmarks and feature/function details on request.

DLSw is based upon an industry standard developed by IBM and placed in the public domain. However, like most standards, there are variations in the marketplace. Many vendors add proprietary extensions that work only with their products. In a mixed-vendor hardware environment, DLSw interoperability will default to a base function set. This usually results in a significant loss of performance and function compared to a single vendor setup. Although most businesses standardize on a single hardware vendor, the performance hit from "base function" needs to be considered if you have a mixed hardware infrastructure.

Additionally, if your routers do not support Advanced Peer to Peer Networking (APPN), scalability may present some problems in large enterprises that choose DLSw. APPN/HPR can offload work from the SNA front ends (3745s) onto cheaper router boxes. DLSw typically adds additional load to the SNA front ends because you get rid of remote 37xx boxes, and so the scalability question boils down to which box you expect to handle the SNA boundary. However, DLSw has inherent scalability limitations because it requires definitions and router resources to increase in proportion to the number of clients.

DLSw users also face the following limitations:

• DLSw requires additional routers on each side of the TCP/IP network in many cases, since only routers support DLSw.

• Heavy workload demands are placed on routers implementing DLSw. They must support intensive TCP connection setup and maintenance involving link-level acknowledgments, TCP retransmit buffers, timers, congestion control logic, protocol translation and store-and-forward processing. This means that higher-end, more expensive routers are usually required.

• DLSw requires significant router configuration. Many routers lack APPN connection network support, which creates a scalability problem. It also results in discovery ("can you reach") messages being forwarded to all partners.

• When one of the DLSw units fails, session outages occur, even if backup routers are in place.

EE, also known as "HPR over IP," is a software-based solution that extends the existing open SNA high performance routing (HPR) technology. EE efficiently integrates the SNA application frames into UDP/IP (User Datagram Protocol/Internet Protocol) packets in the IP stack of the application’s communications server. These packets then travel through the IP network as standard IP packets. UDP/IP packets are the most basic data packets to travel over the IP network and do not require any special hardware or software changes to support their transmission.

To the HPR network, the IP backbone is a logical link. To the IP network, the SNA traffic consists of UDP datagrams that are routed without hardware or software updates to the IP backbone. There is no protocol transformation as there is with gateways. Unlike common tunneling mechanisms, the integration is performed at the routing layers without the overhead of additional transport functions. This enables efficient use of the intranet infrastructure for support of IP-based clients accessing SNA-based data, as well as clients running any of the SNA LU types. EE is supported across the entire line of IBM’s Communications Servers and routing hardware. While software-based, EE can be implemented on routers.

EE offers a number of advantages over a hardware-based solution. First, it minimizes single points of network failure, since it is an "end-to-end" rather than a "node-to-node" solution. No individual router controls the success or failure of the EE transmission. Rather, EE exploits your network "rollover" in the event of a router failure without any additional requirements.

EE also provides better traffic control through class-of-service-based traffic scheduling across TCP/IP networks. It also allows routers in the IP backbone to route traffic, based on a priority scheme derived from UDP port numbers or the IP preference bits. DLSw lacks the ability to provide this level of granularity, treating all SNA traffic (batch and interactive) the same in the TCP/IP network.

EE improves scalability and cost-effectiveness because it doesn’t require nearly as many router resources as DLSw and adds very minimal additional cost to the servers. EE effectively removes all of the TCP/IP overhead and incremental router workload of DLSw. Network reliability functions and congestion management are more cost-effectively managed on the end systems than in the routers. EE reduces the demands on the data center routers and front-end processors by eliminating the need to perform protocol translation and store-and-forward functions associated with the transport layer. This frees increasingly expensive router capacity for more efficient network packet routing.

Without the overhead of DLSw, EE performance is up to 10 times faster in similar network configurations. This is because one-tenth as many resources are consumed at the edge router. Even if routers provide the EE function, only the packet forwarding function is needed. This significantly reduces the processing required at the routers for routing SNA traffic over TCP/IP networks, when compared to other technologies.

Mainframe users of Enterprise Extender receive the following additional benefits that DLSw cannot offer:

Full support for parallel sysplex. This allows applications running in an S/390 parallel enterprise server to move to other processors without terminating the original session, in the event that the original applications processor fails. This results in "five nines" availability of the underlying application (99.999 percent availability) running on OS/390.

Integration of the HPR network all the way to the data center. The TCP/IP network "cloud diagram" no longer ends with an "SNA stick" from a router to the mainframe. With EE functions running on the communication servers, there is no need to purchase and/or install a router that supports SNA.

Virtually no overhead in System/390 when EE is running on OS/390. System/390 features highly optimized processing for IP and SNA functions within a single Communications Server stack. EE traffic travels once through the communications stack – "half" on the SNA side and "half" on the IP side.

New IP functionality released in OS/390 and the Communications Server for OS/390 without rewriting application code. EE allows OS/390 and the network to view an SNA application as a pure IP data stream and manages this data accordingly. Recent performance increases in the OS/390 IP stack have delivered 15X improvements in Web-serving and mixed IP workload.

About the Author: Christopher C. Holley is the Brand Manager for IBM’s eNetwork Communications Server for OS/390 products. He can be reached at chrisho@us.ibm.com.

***********************************************************************

Sidebar:

Consolidating your networking on IP is not a trivial decision. It will change the way you do business and affect the skills needed in your IT department. Be sure to include the hard and soft costs, including the ongoing "people cost." Put a member of the finance group on your evaluation team. Finance can provide valuable insights, and its buy-in will make it easier to get approval from a CIO or business line manager.

Don’t assume that router solutions are always less expensive than mainframe-based solutions. Mainframe MIP costs have decreased substantially over the last several years and will continue to do so. Today, you will find an S/390 solution as price-competitive as any solution on the market.

Every enterprise considering an "SNA over IP" solution should look at both Data Link Switching and Enterprise Extender. In addition to cost, the following factors are crucial to deciding between DLSw and EE: Organizational preference for a hardware or software solution, available skill set in the IT department, available capacity on routers or servers, willingness of vendors to deal, urgency of your IP consolidation timetable, location and role of TN3270 server, future network plans and workloads, customer references in relevant industries, and interoperability.

– C.H.