Migration of a software application from one platform (32bit OS1 on xxx hardware) to another platform (64bit OS2 on yyy hardware)
Upgrade of a software application involves upgrading the software from a lower version to a higher version (like XX version of database to YY version of database)
Both the above activities are technical and no business benefit is directly attributed to this type of activity on the software applications.
An application enhancement will involve providing a new feature or functionality to the end-user of the application. Sometimes when the application is majorly enhanced, the new enhanced application can get migrated to a new platform and/or the underlying technology stack upgraded to higher version.
The key question I am dealing in here is:
When there is no "enhancement" involved, purely a migration and/or upgrade is planned:
a. is it good to do both together in a big bang? OR
b. is it good to migrate first to the new platform and upgrade the software later? OR
c. is is good to upgrade the software first and then migrate to new platform?
a. One can do it together but if there is an issue it is difficult to find the cause of the problem due to new platform or new software version.... each can point fingers at the other....
b. is a good choice if the platform migration is more complex than software upgrade.
c. is a good choice if the software is more stable on the higher version and if any specific platform related issues are fixed in the higher version.
If the software application has multiple tiers with database, application server etc., one should carefully choose the overall path of migrate and upgrade. Making sure the impact on the business SLAs is minimized; It is also important to have a very good "back-out" plan, if something badly goes wrong!!
Tuesday, March 23, 2010
Thursday, March 11, 2010
Performance Analysis & Tuning
Performance of any system is defined as "accomplishment of a 'task' measured against agreed quality, cost and time."
The software tasks are primarily divided into "transactional" tasks and "analytical" tasks.
One big area in Software Systems management is the Performance Analysis and Tuning.
How to measure the performance?
The basic thing needed for "Performance" is the "Measurement" - What is to be measured and How?
Generally people measure the "response time" (time taken to complete the task) if the task is transactional
and
throughput (number of tasks performed in a fixed unit of time) if the task is analytical.
So, in both the cases what is measured is "time". What happened to the original definition that consisted the "quality" and "cost"?
In Software systems the "tasks" are automated and once the automation is tested well, "quality" is guaranteed for the task.
The "cost" of performing the automated task of a software application is the computing resources (CPU and memory) that are allocated to the application and how well the application can use additional resources to carryout the tasks.
Assuming the application can take full advantage of available CPU and Memory resources allocated to it, "TIME" becomes the best measure to analyze the performance of a software system.
How to Tune the system?
Reduce the amount of time taken to perform the 'task'. As simple as that.
Is it that simple?
Let us take a close look at this approach. Can we really reduce the time taken to perform a task?
Answer is "NO". If a task takes "x" amount of seconds to do, it takes "x" seconds to finish it.
But,
Every Software task has two components of this "time". What are they?
a. The time taken to actually DO the work
b. The time taken while waiting for a dependent sub-task to get completed.
So, Now if we ask the same question Can we reduce the time taken to perform a task?
The answer is YES - by reducing the time "waiting"!
Performance Tuning is all about scheduling and executing the sub-tasks of a "task" optimally such that the waiting time is minimized (if not completely removed) thereby using the available resources optimally and reducing the "cost" of performing the "task" That is the secret of performance tuning!
The software tasks are primarily divided into "transactional" tasks and "analytical" tasks.
One big area in Software Systems management is the Performance Analysis and Tuning.
How to measure the performance?
The basic thing needed for "Performance" is the "Measurement" - What is to be measured and How?
Generally people measure the "response time" (time taken to complete the task) if the task is transactional
and
throughput (number of tasks performed in a fixed unit of time) if the task is analytical.
So, in both the cases what is measured is "time". What happened to the original definition that consisted the "quality" and "cost"?
In Software systems the "tasks" are automated and once the automation is tested well, "quality" is guaranteed for the task.
The "cost" of performing the automated task of a software application is the computing resources (CPU and memory) that are allocated to the application and how well the application can use additional resources to carryout the tasks.
Assuming the application can take full advantage of available CPU and Memory resources allocated to it, "TIME" becomes the best measure to analyze the performance of a software system.
How to Tune the system?
Reduce the amount of time taken to perform the 'task'. As simple as that.
Is it that simple?
Let us take a close look at this approach. Can we really reduce the time taken to perform a task?
Answer is "NO". If a task takes "x" amount of seconds to do, it takes "x" seconds to finish it.
But,
Every Software task has two components of this "time". What are they?
a. The time taken to actually DO the work
b. The time taken while waiting for a dependent sub-task to get completed.
So, Now if we ask the same question Can we reduce the time taken to perform a task?
The answer is YES - by reducing the time "waiting"!
Performance Tuning is all about scheduling and executing the sub-tasks of a "task" optimally such that the waiting time is minimized (if not completely removed) thereby using the available resources optimally and reducing the "cost" of performing the "task" That is the secret of performance tuning!
Subscribe to:
Posts (Atom)
