This white paper provides a brief description of the requirements and the hurdles of managing performance for Linux under z/VM. For a list of testimonials gathered from the listserv, click on customer testimonials.
There are four functional requirements for most installations to manage Linux under z/VM, as well as a large requirement for access to skills. Velocity Software's ZVPS is recommended by hundreds of installations because of ZVPS's ability to meet the functional requirements as well as Velocity Software's ability to provide performance analysis skills to solve customer problems that are rarely intuitive.
Velocity Software's product focus is a complete solution to meet the requirements - and to fit into an installations existing framework. For example, Velocity Software provides data as input to MICS from CA or MXG from Merrill Associates, and provides operational alerts that can be displayed on an HP Openview console. Allowing the Linux on z/VM platform to participate in the enterprise allows all job functions to support this new environment with little change.
The Functional requirements for managing Linux under z/VM are fully met by ZVPS and include:
One of the major requirements in a shared resource environment is to have the ability to charge users for resources consumed. There are two parts of this requirement in a Linux under z/VM environment, one for z/VM, and the other at the Linux process level or Linux application level. The traditional z/VM charge back model was to charge by virtual machine, by CPU second consumed, memory page resident, and DISK I/O. Click here for a short white paper discussing this data.
A second accounting requirement for Linux is at the process level. Though data is captured at the process level, it is more important to group the process data by application. ZVPS provides multiple mechanisms for doing this, with all the data grouped by Websphere application, or in several different user tailorable methods. Click here for a short white paper discussing the data available.
Understanding the infrastructure issues and how to manage the costs for collecting accurate data makes the Linux under z/VM platform much more attractive. However unprepared installations find themselves with data that is not useful, or costs significant resources to collect, should it be collected - and in this case often must disable key parts of the management infrastructure. One vendor was heard to say "only measure when there is a problem", suggesting a total lack of ability to manage the environment, but that is their offering.
There are infrastructure requirements that may not be intuitive to the new installation. There is a requirement for correct data, complete data, and captured at a low cost. For installations that commonly have 100 or more Linux servers running in a single LPAR, the cost of collection from each server becomes a major consideration. If the cost of instrumentation for each server is 5% of a processor, then these 100 servers will consume 5 IFLs just for instrumentation. Discussed like this, very few would want this configuration, however, there are management tools still being marketed that have have this problem. Velocity Software's published target for data collection is .1% of a processor per node. Thus the cost of FULLY instrumenting 100 servers is normally about 10% of ONE CPU.
The data provided by Linux in a virtual environment is wrong. This has been published many many times, with examples easily showing CPU data collected from inside Linux wrong by an order of magnitude. Most vendors will not discuss this problem (or mumble when asked). Data that is this far wrong can never be trusted - a process that is shown to be using 50% of a CPU may be perceived to be a problem, but in reality, the process maybe using less than 5% in which case it is normal. With this kind of innacuracy, this data can never be depended upon for performance analysis, capacity planning, chargeback, or even operational alerts. Velocity Software was the first to publish this research and provides the ONLY product that corrects the data - correcting the LInux data based on the highly accurate z/VM data.
There are MANY issues with managing virtual linux servers.
As Velocity Software is highly active in research with many
large Linux installations, problems are found and corrected
long before our customers even recognize the problem.
Many of these issues are irrelevant to small installations.
During small pilots, installations not aware of the issues
will not get an understanding of the true cost of production.
At the point where you run 20 or more linux servers, avoiding
these problems becomes financially critical.
Installations that do not create a proper infrastructure for
operating Linux under VM will fail.
The expense of poor
performance and capacity planning has already led to many
project failures. Proper infrastructure is required for
successful consolidation projects.
Integrated Data
As ZVPS maintains all of the data in a very integrated
format, many issues are easily resolved in just the processing
of the data. The central data repository allows data analysis
without impacting local servers and provided the ability to
show an integrated picture.
Valid CPU data:
As Linux servers may run many processes, data on
each process is required for performance analysis and capacity
planning. Linux reports all of it's process numbers based on
what Linux perception is at the time. If the system utilization
is very high, Linux perception is that Linux must be using
all of the resource - and reports based on that information.
Thus the Linux process numbers reported by any linux monitor
may be highly exagerated. A factor of 10 on a busy system
has been presented by Velocity Software at many seminars
and conferences. And the fact that Velocity Software has a
unique methodology of correcting this data not provided by
any other vendor.
Why have the expense of running a monitor
if the data provided is wrong and not useful?
Cross platform at no additional cost:
Because the standard interfaces provided using SNMP, and
NETSNMP - an open source package incorporated by Redhat and Suse
there are few support issues. NT, Linux, Sun, HP all have
implementations of this data.
Thus at no extra expense, an
installation using ZVPS may monitor many different platforms,
and many different nodes without having to purchase unique
agents to run on those platforms.
Instrumentation cost:
In terms of resource requirements, running an agent on 100's of
different virtual servers will have a large and measureable
cost. The cost of running NETSNMP is measured and often
presented by Velocity Software at conferences and seminars as
being VERY low. Much less than 1% per server. As a cost of your
infrastructure, you MUST understand the agent requirements. If
the cost of running the agent is too high, then you can not
afford to run the agent.
Why use a performance monitor that you
must turn off to correct the performance problem???
Before
committing to an agent, measure the cost, and multiply that by
the number of servers you expect to operate. Can you afford to
operate it?
Supported Releases:
One added requirement is full and current support. Velocity Software
currently supports all versions of SLES that operate on "z", as well
as all versions of RHEL that operate on "z". Other distributions have
not been requested but will be supported as necessary. All release
of z/VM are supported with current support including all releases of z/VM
including z/VM 5.3.
Summary
In building an infrastructure for a Linux running under z/VM these are the questions to answer:
Can the following requirements be met:
and with the following requirements:
In addition, in selecting a vendor, ensure the vendor has the skills you will need available to you to resolve your problems. There will be problems, and with MOST published performance research authored by Velocity Software employees, most installations fully appreciate the skills that can be applied to customer problems.
ZVPS and Velocity Software are by far the leaders in solving your performance and capacity planning issues. With an integrated architecture that is easily extensible, ZVPS will always stay the leader in helping installations meet the needs of their virtual linux farm.
