Velocity Software, Inc. is recognized as a leader in the performance measurement of z/VM and Linux on z. The Velocity Performance Suite consist of a set of tools that enable installations running z/VM to manage Linux and z/VM performance. In addition, many components of server farms can be measured and analyzed. Performance data can be viewed real-time through the use of either 3270 or a browser. The CLOUD Implementation (zPRO) component is designed for full cloud PaaS implementation as well as to extend the capabilities of the z/VM sysprog (system programmer) to the browser world. This feature moves system management to the point-and-click crowd. Archived data and reports can be kept available of long term review and reporting usine zMAP. The zVPS, formally ESALPS, components consist of: zMON (formally ESAMON - real-time display of performance data), zTCP (formally ESATCP - SNMP data collection), zMAP (formally ESAMAP - historical reporting and archiving), zVWS (formally ESAWEB - z/VM based web server), zTUNE (a subscription service), zVIEW (formally SHOWCASE - web based viewing of performance data), zPRO (new to the quality line of Velocity Software Products). Velocity continues to work with other software vendors to ensure smooth interface with or from other products such as VM:Webgateway, CA-Webgateway, EnterpriseWeb, MXG, MICS. Velocity software remains the leader and inovator in the z/VM performance, Linux performance, Managing cloud computing arenas.
About Us | Products | FAQ | Demo | Customer Area | Education | Linux Hints & Tips | Presentations | News | Industry and Events | Employment Opportunities
Home | Contact Us | License Info | Newsletter    

Linux Process data is Wrong

Installations that expect to use "top" or other agents running in Linux under z/VM to determine capacity requirements or current CPU useage should know that the data is VERY wrong. All agents under Linux get their data from exactly the same /proc table, and thus all agents have exactly the same problem. The means of CPU accounting for Linux is to wake up 100 times per second and allocate the 10ms of CPU based on which process happens to be using the CPU at that time. But in a virtual environment, Linux doesn't always get dispatched at exactly the 10ms mark, and doesn't have the ability to recognize this fact. When the VM system is very busy, Linux is not dispatched consistently, and does accounting for processor a bit randomly.

This short presentation shows a very typical case using real data, and shows how Velocity Software has the technology to provide signficantly better data that can be used for capacity planning and performance analysis. This example shows the process data as being reported at 10 times the actual value. Cases much worse than this are common. Imagine if you have one process in a loop, and all your z/VM processors are 100% utilized, with 20 linux servers. In this case, every server you use "top" to measure will show it is using 100%. Which server should you eliminate to solve the problem?

The solutions to this problem are to
1) use the VM monitor to determine which Linux server is using the CPU, and then use "top" to determine the process using the CPU. This is difficult, time consuming, and does not lend itself to capacity planning or timely problem resolution.
2) Fix Linux process accounting to use the CPU timer instead of the TOD timer, will not happen in near future.
3) Use Velocity Software's ESALPS technology to provide accurate data...


Click here to start

Table of Contents

1) Performance Instrumentation Issues
2) Performance Case Study
3) Performance Case Study
4) Performance Case Study

How ESALPS provides accurate data.
5) Prorate Linux data
6) Prorate Linux data
7) Prorate Linux data

Why ESALPS can fix the data?
8) Prorate Linux data
9) Prorate Linux data
10) Prorate Linux data
Author: Barton Robinson
Mail us



Don't miss Velocity Software's Performance Seminars