20 Experiences with Oracle for Linux on zSeries
The topology of a network using IUCV is the same as that illustrated in Figure 2-2
on page 18 using CTC adapters.
We have found that Guest LAN is easier to manage and provides better
performance, so we recommend that you use the Guest LAN facilities of z/VM
Version 4.3 rather than point-to-point connections with IUCV.
Network Topology without z/VM
If you deploy a number of Linux machines in logical partitions, without the use of
z/VM, then each Linux machine will be a peer on the network with the other
operating systems running in other logical partitions. This is illustrated in
Figure 2-4. Note that unlike a virtual machine under z/VM where the network
adapter address (NIC) has the same address as other virtual machines, when
you deploy in a logical partition, the operating system must see the real device
address of the NIC. Each of the Linux machines in Figure 2-4 will use different
device numbers to access the OSA/Express adapter.
Figure 2-4 Peer Linux guests in logical partitions
2.1.5 Capacity planning
Deploying Linux on zSeries processors must be approached with the same
methodology as deploying any other operating system. Consolidating multiple NT
or Linux servers onto Linux for zSeries can be a great idea, and can save money.
However, if you are planning to consolidate servers that are doing existing
MVS01z/OS z/OS z/OSz/OS
Shared OSA/Express LAN
Linux Linux Linux
Chapter 2. How to deploy Oracle9i on Linux/390 21
workloads, then you must measure the utilization of the existing servers and also
do some calculations to ensure that the configuration of Linux on the zSeries
processor is appropriate. The ITSO Redpaper
Server consolidation with Linux for
REDP0222, covers many of the capacity planning issues.
Performance considerations for Linux on zSeries are discussed more completely
in Chapter 8., “Performance and scalability considerations” on page 131. This
section will only discuss the basic considerations.
Number of CPs
The speed and capacity of a z/800 or z/900 machine are well understood.
However, deploying Linux on these machines will probably share these
resources through the use of LPAR with other workloads and other operating
systems. It will be fairly rare to dedicate a complete zSeries machine to a single
copy of Linux, even though that is the norm in other platforms. Each of the small
servers that are being consolidated onto the zSeries processor has the entire
capacity of the server available for peak loads, but probably (hopefully?) runs at
an average utilization much less than 100%. On the zSeries machine, all
resources are shared, and hopefully the peak activity of each Linux machine will
not occur simultaneously. You must analyze each of these servers and calculate
the expected utilization and number of CPs required to provide the capacity
There are tools available from IBM to assist in this process. For a detailed sizing
study using a tool called SIZE390, contact your IBM representative. This tool can
help analyze the utilization of the servers that will be consolidated onto the
As discussed in “Deploying in a logical partition” on page 12, each logical
partition is assigned a
weight. Processor resources are given to the various
logical partitions in the ratio of the weights. Normally, we recommend that the
sum of the weights assigned to the logical partitions sum to 100 or 1000. If this is
done, then the weight can be directly understood to be the percentage of total
capacity of the zSeries processor that is given to the specific logical partition.
Capacity greater than this percentage can be delivered to a given logical
partition, but only if no other logical partition wants the resources. The weight,
expressed as a percentage is the
target capacity of the logical partition. If, for
example, you want the Linux machines to get 40% of the capacity of the zSeries
processor, then the sum of the LPAR weights for the Linux logical partition should
be 40% of the total weights for all logical partitions.