When Do I Upgrade My Processor?

When Do I Upgrade My Processor?
Performance Analysis with Utilization Software

Most computer users become aware of the inadequacies of their computer when it appears to slow down. On-line response time may be getting longer between transactions or it may be moving to the limit of tolerance. Batch jobs may be taking more time than they used to or maybe nighttime runs interfere with the daytime on-line work. There are times when these situations happen quickly, but most often, it is a gradual process.
The things that cause a computer system to be degraded are not always a lack of processor power. Memory, disk, and data communications need to be considered. In addition, the system has factors, buffers, and many options that can be adjusted to improve performance. DMSII users have several additional issues to consider from what other users need to examine.
It is generally not a good idea to have the hardware vendor, who wants to sell another processor to you, evaluate your current system. The vendor's bias may not permit a thorough investigation into the causes of current degradation or bottlenecks. An independent analysis is always a wise decision. Using the technical expertise of a disinterested third party will usually provide a more thorough and in-depth study of your computer system.
Since there are so many issues to consider, jumping to the conclusion that a bigger processor is needed may be a costly mistake. There are many things to consider when a computer is experiencing degradation, be it temporary or permanent. I could tell you about an A5F user who moved to an A6K dual processor system with double the processor and double the memory of the A5K and saw response time more than double. That is, it did not decrease, but got much worse. Once you spend the money, it is too late to go back. Performance analysis is needed on all computer systems, but capacity planning is a pro-active means to apply performance analysis to a business plan and make decisions for the future when it will help most rather than making a panic decision when one is pressured by the wrong people and possesses inadequate information.
Capacity planning should start with the installation of a computer. First a base line is established. Then, performance information is gathered regularly. How often data should be gathered is determined by many site circumstances. One may gather performance data each week or month, but at least once a month. A business plan must be established and updated over time. An experienced performance analyst can use the business data and performance data as well as a knowledge of the applications running on your computer to make reasonable estimates of when you will need to increase or upgrade components of your computer system. There may come a time for a whole new replacement, but usually needed resources can be updated to keep you operating at a reasonable performance level with a proper weight of economical considerations.
Performance analysis requires software tools, programs, to gather utilization data and then display it in useful graphs and reports so that it may be analyzed. Most performance software, such as SMFII, has never been accused of being user friendly. They are truly hard to read and understand. Experienced performance analysts are required to accurately interpret the data and make recommendations. It has been my attempt to make useful utilization reports available in my own performance software. My recommendation is that every site should have performance software that creates easily understandable reports for people with limited computer knowledge.
Given that you have user friendly performance software and you have had a performance analyst perform a base line analysis for you; next, you are ready to start your own weekly or monthly performance monitoring. With the proper reports, you can watch your computer grow and know what resources are growing most. As long as you are not reaching a danger point (i.e. out of memory) condition you should be able to monitor your own system. When you have questions or concerns, you should have someone to call that you can rely on. Once every six to twelve months you should have a performance specialist review your data to verify that you are progressing properly. There may be a need for some adjustments at these intervals.
The point is that you, the user, can control your own performance analysis and use a specialist only when you have a need. You should have a regular plan and compare your system growth with the plan. Are you on course or do you need to make changes to accomplish your goals. A specialist can make recommendations as to when you will need to consider adding or upgrading resources so that you can make the needed economical and operational plans.
You, the owner and user of the computer, should be in charge of how you use your computer and where you are going with it. Don't let the computer and its supposed needs dictate to you how and what you are doing with it. There will be areas of expertise that you will not have from time to time. Establish ahead of time who you will use and trust in these areas. Use people resources when needed. Don't let your computer operations control you. You take charge of you computer installation and get help when you need it.
Modeling is another issue that is often used in performance analysis and capacity planning. It is a way of taking your present and forecasted workloads and having a modeling person or program recommend what resources you need or will need. I have seen many forecasts by what has been called the premier modeling program in the computer industry. It has proved helpful on occasion.
Most performance analysts are in favor of using modeling programs. This is an area where I disagree. It is a known fact that when the best modeling program gives its best results it has a 25% error factor. More often than not this error rate is much higher than 25%. It could be 50% or more. I believe that there are a few people in the computer industry who can out-perform modeling programs. I think that I can say without boasting that I have not made any recommendations in forecasting that have come as high as a 25% error factor. I know others who can do the same.
Most any technical person that Unisys has can do a "SMFII analysis" and tell you that you need a bigger processor, maybe even the same one that the salesman wants to sell you. My experience has been that a "SMFII analysis" is rarely a true performance analysis.
A performance analysis is not geared to selling hardware, although that may take place. It is directed at revealing the true measurement of your computer system and giving steps to correct or improve the situation. A capacity study or plan will further indicate a forecast of where your computer and its resources are headed and compare that with a business plan. Recommendations will then be made on a planned course for a given period of time. Performance studies should be made along the way to determine whether or not you are following your business plan and where adjustments are needed.
The bottom line is that you, the computer user, can chart your own future with your A Series computer. With a little help from a performance specialist you can have confidence in your direction with computer resources. With guidance and good software you can do most of the work. You must understand how the A Series computer operates. A brief presentation or seminar can give you the needed information if you do not already have it.

Performance Products
As I travel to different A Series sites I have observed different software performance products. The PC monitors and graphical information are really impressive. I was even present at a Teamquest presentation of the new SMFII features. Unisys markets SMFII which is a product from Teamquest. One of the better products that I have seen was at a European site.
After seeing what others are producing over the last 20 years (I first entered the performance arena in 1976), I wonder how much of the software is written for show instead of analysis benefit. What I mean by that is that the displays and graphs are attractive to the senses, but functional analysis is needed for proper performance analysis and capacity planning.
Over the past 10 to 15 years I have developed a product that I use for my own purposes. I call it Utilization Software. It has evolved as I see new needs for my analysis. It is certainly not as fancy as most software on the market today, but it was not developed to be marketed. It was developed as a tool to give the best analysis that I think can be done.
Although Integrity Services now markets the utilization software, it was developed with performance analysis in mind, not marketing. This is probably the one thing that I have noticed with today's performance software - it is written to impress the user rather than do a complete job of performance analysis. Even where certain reports are available, they must be configured; half-hour summaries are impossible or hard to get. The best tools should be load and go. The default reports should require very little if any adjustments to parameters.
Now I realize that this is my view of performance analysis and performance software tools and products; but it is my professional opinion. A good performance tool is going to have critical information in summary form at regular intervals that graphical representation will not be able to properly portray. Examples would be disk queuing and throughput rates (number of bytes/second and number of I/Os) and mix reports at regular intervals sorted by resource use such as CPU. I have found 30 minute intervals to be very effective in performance analysis.
On-line resource evaluation is also a valuable tool. When certain resources are in use beyond specified thresholds the operator is informed and steps may be taken to remedy the situation. If thresholds are set properly then enough time will be allowed to adjust system patterns before they adversely affect the system and its users. An example would be that if you knew that a disk resource was exceeding a threshold, you could attend to it before an out of disk situation occurred. In the same way when processor or memory resources were being used beyond a given threshold, then adjustment could be made before a "thrashing" situation occurred.

UTILIZATION STATISTICS
(For NX and A Series Computers)
Setup is not required. Just load the files and start the job as follows:
COPY = FROM UTILIZATION
START JOB/U (240)...where 240 minutes is the desired sample time.
The job may also be run under a usercode, if desired.
Of the following available reports, 1, 2, 3, 5, and 6 are listed in the above job:
1. Summary information
2. Processor and I/O detail utilization statistics graph
3. Detail memory usage statistics graph
4. Unit queue detail report (For specified I/O units)
5. Subtotal summary information
6. Sorted mix report every [x] minutes
A selected time period may also be specified for the reports.
The first program is U/COLLECTOR. It gathers all the necessary information and stores it in MEASUREMENTS, ENVIRONMENT, and UNITSTATS (Disk files). The internal value supplied to collector is a time value (in minutes) that it is to collect information (i.e. RUN U/COLLECTOR ("MIX 15"); VALUE=30 will run the program for 30 minutes). A sorted mix will be produced every 15 minutes. Also, any accept message will cause collector to go to EOT. The second program is U/STATISTICS. It produces reports based on input specifications. The following parameters are used at run time for U/STATISTICS:
Parameter Values
Summary only (default)... Blank
Specify report times ............. RMT
Processor utilization.............. CPU
Memory usage ..................... MEM
Detailed unit queue report.... UR
Print a mix report.................. MIX ............ (The frequency of the mix is specified to U/Collector)
Skip n samples (overview).... FREQ n
Subtotal every n samples...... SUBT n
Sort unit report by worst case WORST n... (n = number of units)
Print disk only in unit report.. DISK
A few examples should illustrate the use of parameters:
RUN U/STATISTICS("CPU MEM"): Processor, memory and summary reports
RUN U/STATISTICS("CPU RMT"): Processor and summary reports for the specified time
period. (Times are entered remotely at a terminal)
RUN U/STATISTICS("MEM SUBT 30"): Memory, subtotals for 30 minutes, and summary
reports
RUN U/STATISTICS("CPU FREQ 60"): Processor report every 60 minutes
Note 1: Subtotal and total reports include statistical reports showing average use of processor in
eight different categories, IO rates and activities, ready queue activity, memory usage, unit
queuing and IO rate separated into reads and writes, and system environment reports.
Note 2: Unit numbers for detailed unit reports may be entered via a job file or remotely depending
on the place of execution.
Executable program code and a JOB file with default settings are provided.
Specify: 4.3 or 4.4 or 4.5 Release


When Do I Upgrade My Processor? Performance Analysis with Utilization Software Most computer users become aware of the inadequacies of their computer when it appears to slow down. On-line response time may be getting longer between transactions or it may be moving to the limit of tolerance. Batch jobs may be taking more time than they used to or maybe nighttime runs interfere with the daytime on-line work. There are times when these situations happen quickly, but most often, it is a gradual process. The things that cause a computer system to be degraded are not always a lack of processor power. Memory, disk, and data communications need to be considered. In addition, the system has factors, buffers, and many options that can be adjusted to improve performance. DMSII users have several additional issues to consider from what other users need to examine. It is generally not a good idea to have the hardware vendor, who wants to sell another processor to you, evaluate your current system. The vendor's bias may not permit a thorough investigation into the causes of current degradation or bottlenecks. An independent analysis is always a wise decision. Using the technical expertise of a disinterested third party will usually provide a more thorough and in-depth study of your computer system. Since there are so many issues to consider, jumping to the conclusion that a bigger processor is needed may be a costly mistake. There are many things to consider when a computer is experiencing degradation, be it temporary or permanent. I could tell you about an A5F user who moved to an A6K dual processor system with double the processor and double the memory of the A5K and saw response time more than double. That is, it did not decrease, but got much worse. Once you spend the money, it is too late to go back. Performance analysis is needed on all computer systems, but capacity planning is a pro-active means to apply performance analysis to a business plan and make decisions for the future when it will help most rather than making a panic decision when one is pressured by the wrong people and possesses inadequate information. Capacity planning should start with the installation of a computer. First a base line is established. Then, performance information is gathered regularly. How often data should be gathered is determined by many site circumstances. One may gather performance data each week or month, but at least once a month. A business plan must be established and updated over time. An experienced performance analyst can use the business data and performance data as well as a knowledge of the applications running on your computer to make reasonable estimates of when you will need to increase or upgrade components of your computer system. There may come a time for a whole new replacement, but usually needed resources can be updated to keep you operating at a reasonable performance level with a proper weight of economical considerations. Performance analysis requires software tools, programs, to gather utilization data and then display it in useful graphs and reports so that it may be analyzed. Most performance software, such as SMFII, has never been accused of being user friendly. They are truly hard to read and understand. Experienced performance analysts are required to accurately interpret the data and make recommendations. It has been my attempt to make useful utilization reports available in my own performance software. My recommendation is that every site should have performance software that creates easily understandable reports for people with limited computer knowledge. Given that you have user friendly performance software and you have had a performance analyst perform a base line analysis for you; next, you are ready to start your own weekly or monthly performance monitoring. With the proper reports, you can watch your computer grow and know what resources are growing most. As long as you are not reaching a danger point (i.e. out of memory) condition you should be able to monitor your own system. When you have questions or concerns, you should have someone to call that you can rely on. Once every six to twelve months you should have a performance specialist review your data to verify that you are progressing properly. There may be a need for some adjustments at these intervals. The point is that you, the user, can control your own performance analysis and use a specialist only when you have a need. You should have a regular plan and compare your system growth with the plan. Are you on course or do you need to make changes to accomplish your goals. A specialist can make recommendations as to when you will need to consider adding or upgrading resources so that you can make the needed economical and operational plans. You, the owner and user of the computer, should be in charge of how you use your computer and where you are going with it. Don't let the computer and its supposed needs dictate to you how and what you are doing with it. There will be areas of expertise that you will not have from time to time. Establish ahead of time who you will use and trust in these areas. Use people resources when needed. Don't let your computer operations control you. You take charge of you computer installation and get help when you need it. Modeling is another issue that is often used in performance analysis and capacity planning. It is a way of taking your present and forecasted workloads and having a modeling person or program recommend what resources you need or will need. I have seen many forecasts by what has been called the premier modeling program in the computer industry. It has proved helpful on occasion. Most performance analysts are in favor of using modeling programs. This is an area where I disagree. It is a known fact that when the best modeling program gives its best results it has a 25% error factor. More often than not this error rate is much higher than 25%. It could be 50% or more. I believe that there are a few people in the computer industry who can out-perform modeling programs. I think that I can say without boasting that I have not made any recommendations in forecasting that have come as high as a 25% error factor. I know others who can do the same. Most any technical person that Unisys has can do a "SMFII analysis" and tell you that you need a bigger processor, maybe even the same one that the salesman wants to sell you. My experience has been that a "SMFII analysis" is rarely a true performance analysis. A performance analysis is not geared to selling hardware, although that may take place. It is directed at revealing the true measurement of your computer system and giving steps to correct or improve the situation. A capacity study or plan will further indicate a forecast of where your computer and its resources are headed and compare that with a business plan. Recommendations will then be made on a planned course for a given period of time. Performance studies should be made along the way to determine whether or not you are following your business plan and where adjustments are needed. The bottom line is that you, the computer user, can chart your own future with your A Series computer. With a little help from a performance specialist you can have confidence in your direction with computer resources. With guidance and good software you can do most of the work. You must understand how the A Series computer operates. A brief presentation or seminar can give you the needed information if you do not already have it. Performance Products As I travel to different A Series sites I have observed different software performance products. The PC monitors and graphical information are really impressive. I was even present at a Teamquest presentation of the new SMFII features. Unisys markets SMFII which is a product from Teamquest. One of the better products that I have seen was at a European site. After seeing what others are producing over the last 20 years (I first entered the performance arena in 1976), I wonder how much of the software is written for show instead of analysis benefit. What I mean by that is that the displays and graphs are attractive to the senses, but functional analysis is needed for proper performance analysis and capacity planning. Over the past 10 to 15 years I have developed a product that I use for my own purposes. I call it Utilization Software. It has evolved as I see new needs for my analysis. It is certainly not as fancy as most software on the market today, but it was not developed to be marketed. It was developed as a tool to give the best analysis that I think can be done. Although Integrity Services now markets the utilization software, it was developed with performance analysis in mind, not marketing. This is probably the one thing that I have noticed with today's performance software - it is written to impress the user rather than do a complete job of performance analysis. Even where certain reports are available, they must be configured; half-hour summaries are impossible or hard to get. The best tools should be load and go. The default reports should require very little if any adjustments to parameters. Now I realize that this is my view of performance analysis and performance software tools and products; but it is my professional opinion. A good performance tool is going to have critical information in summary form at regular intervals that graphical representation will not be able to properly portray. Examples would be disk queuing and throughput rates (number of bytes/second and number of I/Os) and mix reports at regular intervals sorted by resource use such as CPU. I have found 30 minute intervals to be very effective in performance analysis. On-line resource evaluation is also a valuable tool. When certain resources are in use beyond specified thresholds the operator is informed and steps may be taken to remedy the situation. If thresholds are set properly then enough time will be allowed to adjust system patterns before they adversely affect the system and its users. An example would be that if you knew that a disk resource was exceeding a threshold, you could attend to it before an out of disk situation occurred. In the same way when processor or memory resources were being used beyond a given threshold, then adjustment could be made before a "thrashing" situation occurred. UTILIZATION STATISTICS (For NX and A Series Computers) Setup is not required. Just load the files and start the job as follows: COPY = FROM UTILIZATION START JOB/U (240)...where 240 minutes is the desired sample time. The job may also be run under a usercode, if desired. Of the following available reports, 1, 2, 3, 5, and 6 are listed in the above job: 1. Summary information 2. Processor and I/O detail utilization statistics graph 3. Detail memory usage statistics graph 4. Unit queue detail report (For specified I/O units) 5. Subtotal summary information 6. Sorted mix report every [x] minutes A selected time period may also be specified for the reports. The first program is U/COLLECTOR. It gathers all the necessary information and stores it in MEASUREMENTS, ENVIRONMENT, and UNITSTATS (Disk files). The internal value supplied to collector is a time value (in minutes) that it is to collect information (i.e. RUN U/COLLECTOR ("MIX 15"); VALUE=30 will run the program for 30 minutes). A sorted mix will be produced every 15 minutes. Also, any accept message will cause collector to go to EOT. The second program is U/STATISTICS. It produces reports based on input specifications. The following parameters are used at run time for U/STATISTICS: Parameter Values Summary only (default)... Blank Specify report times ............. RMT Processor utilization.............. CPU Memory usage ..................... MEM Detailed unit queue report.... UR Print a mix report.................. MIX ............ (The frequency of the mix is specified to U/Collector) Skip n samples (overview).... FREQ n Subtotal every n samples...... SUBT n Sort unit report by worst case WORST n... (n = number of units) Print disk only in unit report.. DISK A few examples should illustrate the use of parameters: RUN U/STATISTICS("CPU MEM"): Processor, memory and summary reports RUN U/STATISTICS("CPU RMT"): Processor and summary reports for the specified time period. (Times are entered remotely at a terminal) RUN U/STATISTICS("MEM SUBT 30"): Memory, subtotals for 30 minutes, and summary reports RUN U/STATISTICS("CPU FREQ 60"): Processor report every 60 minutes Note 1: Subtotal and total reports include statistical reports showing average use of processor in eight different categories, IO rates and activities, ready queue activity, memory usage, unit queuing and IO rate separated into reads and writes, and system environment reports. Note 2: Unit numbers for detailed unit reports may be entered via a job file or remotely depending on the place of execution. Executable program code and a JOB file with default settings are provided. Specify: 4.3 or 4.4 or 4.5 Release

Copyright 2003© Integrity Services, Inc. All rights reserved