Minimizing the routing responsibility of z/OS Communications Server

OMPROUTE may be run on z/OS® Communications Server for a variety of reasons. If the z/OS Communications Server host is being used as an application or server host and the routing daemon is being run primarily to provide access to network resources, or to provide network resources access to the z/OS Communications Server host, then care must be taken to ensure that the z/OS Communications Server host is not overly burdened with routing work. Unlike routers or other network boxes whose sole purpose is routing, an application host z/OS Communications Server will be doing many things other than routing, and it is not desirable for a large percentage of machine resources (memory and CPU) to be used for routing tasks, as can happen in very complex or unstable networks. In this case the z/OS Communications Server should not be configured as a backbone router, either intentionally or inadvertently. Careful network design can minimize the routing burdens on the z/OS Communications Server application host without compromising the accessibility of z/OS Communications Server resources to the network and vice versa. If care is not taken to minimize the routing work required by the z/OS Communications Server host, OMPROUTE may consume excessive cycles or memory processing huge numbers of routing updates from the network. Or the burden of routing updates may become so large that the z/OS Communications Server cannot keep up because of other workloads on the machine. Because OSPF is heavily timer-driven, this could cause loss of adjacencies and routing problems.

The primary way to reduce the routing burdens on the z/OS Communications Server host is by use of OSPF areas. See step 3 for more information. A z/OS Communications Server application host or sysplex can be placed into a non-backbone area with dedicated routers acting as area-border routers. The area-border routers would advertise the z/OS Communications Server resources to other attached areas (for example the backbone) and would summarize the network outside the local area to the z/OS Communications Server hosts. If possible, this can be further refined to reduce routing protocol traffic by use of interarea route summarization, accomplished in OMPROUTE area-border routers by the RANGE and IPV6_RANGE statements, and in Cisco routers with the area range command. For more information, see z/OS Communications Server: IP Configuration Reference and step 4.

An even further, and ideal, optimization would be to make the area containing the z/OS Communications Server application host or sysplex a stub area. A stub area can be configured such that route summaries (IPv4) or prefixes (IPv6) from other areas are not flooded into the stub area by the area border routers. When this is done, only routes to destinations within the stub area are shared among the hosts. Default routes are used to represent all destinations outside the stub area. The stub area's resources are still advertised to the network at large by the area-border routers. You can use this optimization, sometimes referred to as a totally stubby area, if the following conditions apply to your network:
  • It is acceptable to use default routes to reach destinations outside the stub area. This means that either there is only one area-border router connecting the stub area to the rest of the network, or if there are multiple such connections they are redundant, so that it does not matter which one is used to get outside the stub area.
  • You have no non-OSPF destinations to advertise to the network at large. Stub areas do not permit importation of OSPF external routes. This means for example that you do not have a RIP network attached to the stub area, or if you do, you do not want its destinations reachable from the stub area. Other types of routes that cannot be imported into stub areas include direct routes (for example, for networks attached to interfaces that are not running the OSPF protocol) and static routes. RIP or static routes can be used by the z/OS Communications Server that learns them, they just cannot be advertised.
    Tip: If you define your VIPAs as OSPF interfaces in your OMPROUTE configuration file, routes to their addresses will be considered OSPF routes, and therefore importable into the stub area and able to be advertised by the area-border routers to the network at large.
It is highly recommended to put z/OS Communications Server application hosts or sysplexes into stub areas if at all possible.

A further optimization is to prevent z/OS Communications Server from becoming the designated router on multiaccess media, when pure routers that can perform this function are present. On a multiaccess medium, the designated router and the backup designated router will carry the majority of the routing protocol load for all hosts on the medium. While z/OS Communications Server is capable of performing this role, it does impose additional routing overhead on the system. It would be preferable to allow pure routers to perform this role, if they are available. This is accomplished by ensuring that the pure routers' interfaces onto the medium have higher ROUTER_PRIORITY values than the z/OS Communications Server interfaces on the same medium. However, if the only hosts on a medium are z/OS Communications Server, such as in HiperSockets communications, then one or two of them will have to be designated router or backup designated router.

Tip: You can use the IBM® Health Checker for z/OS to check whether the total number of indirect routes in a TCP/IP stack routing table exceeds a maximum threshold. When this threshold is exceeded, OMPROUTE and the TCP/IP stack can experience high CPU consumption from routing changes. For more information about IBM Health Checker for z/OS, see z/OS Communications Server: IP Diagnosis Guide and IBM Health Checker for z/OS User's Guide.