I need to design a solution to replace some FEPs by Cisco routers with CIP interfaces. The main aspect of this job is choose the appropriate number of routers and the capabilities to support SNA sessions provided by FEPs. I know that the one of the key aspect is the rate and size of transactions.
I would like to know how I can get this informations from mainframe. There is any design guide explain how to define the number and configuration of routers ?
It really depends. Do you know what kind of management software installed on the mainframe? I used NETSPY to manage NCP before. There are some tools to monitor DB2 and CICS.
Thank´s for your response,
In many Cisco documentation the common size of transactions is 40 Bytes IN and 1000 Bytes OUT, and one transaction per minute, so if I divide the # of LUs/60, the result is the number of transaction per second. I will assume this because the customer does not have a tool to monitor this.
I already define the Cisco 7513 with 2 CIP2 and RSP16 to be used for mainframe connection, and now I need to define the routers (number and model) to terminate remote DLSw+ peers (and in the future SNASw).
I read that some variables are relevant: # PUs, #LUs, transaction rate, transaction size, # remote routers. Others variables are less important because of network design (Hierarchical ) and all conections are started by remote router.
I would like to know if exist rules to define how many routers are needed and the capacity to handle PUs, LUs and transaction rate.
Only the transaction rate is revelant and important.
DLSw only keeps track of the number of PU (i.e. number of LLC2 connection and DLSw circuit). It does not keep track of the number of LU. In other words, DLSw uses the same router RP/CPU resource in the following cases:
1. 100 PUs with 2500 LUs and no transaction
2. 100 PUs with 100 LUs and no transaction
The only process that is PU related and consumes router RP/CPU resource is LLC2 keepalive. The router has to Receiver Readys (RRs), which is more or less LLC2 keepalive, when the PU is idle for the period of LLC2 idle-time. The default value for LLC2 idle-time is 10 second. If the router has a lot of PUs on it and consumes a lot of CPU/RP cycle on the process "LLC2 timer", just increase the LLC2 timer to 60 second.
There is no rule/formula between transaction rate and CPU cycle. The TME (Techincal Marketing Engineer) just pump traffic into different routers and record the CPU utilization. As the router uses the same amount of CPU/RP resource on packets with 40 bytes and packets with 1000 bytes, the DLSw white paper below only shows transaction rate vs different router platforms:
Some data of this environment:
- 3000 peers (remote routers)
- 6000 PUs
- 300000 LUs
- 4200 transactions/second
I´m planning to use 2x7200VXR with NSE-1 to terminate all DLSw peers (working with load balance - 1500 peers each) and 4x7513 (RSP16) with 2 CIP2s. 2xCAT6500 will provide LAN switching in Gigabit. In case of fail only one CAT6500, 7200 and 7513 (2 routers) will support all environment.
In your opinion this routers are well sized ?
Thank´s for you assistance.
NSE-1 and NPE-300 have similar performance on DLSw traffic. From the URL in my last posting, I do not think that a 7200(NSE-1) can handle 4200 transaction per second. At least, it is close to 95% to 100% CPU utilization.
Do you consider to use the RSP16s for DLSw termination?
I neeto to explain better, are 4200 transactions/second in mainframe. I think that if I install 2 or 4 7200 this number will be divided, so each router will work with 2100 transactions/second or less.
This is correct ?
I am concerning when one of the 7200s goes out, like a router crash or a maintenance window. The remaining 7200 may not able to handle 4200 transaction/second.
Ok, now you are talking about fault tolerance and redundancy. But you don't think that use many 75xx with a RSP16 and only 2 Gigabit interfaces is a bad cost x benefit relationship ?