Lesson 28. Estimating the cost of 64-bit migration of C/C++ applications
- Purchasing 64-bit hardware and software
- Purchasing a compiler to build 64-bit applications
- Purchasing 64-bit versions of libraries
- Staff training and purchasing additional tools
- Code modification
- Adapting the testing system
- Protection of software units
- Distribution kit adaptation
Starting to plan the port of your project to a 64-bit system, you must be able to estimate the amount of work and material costs. Let us consider those components that make up the total cost of creating a 64-bit software product.
If you find it difficult to estimate the cost of the move to 64-bit systems, you may contact experts of our company OOO "Program Verification Systems" for advice. Our company can also take a part of or all the work of adapting your projects for 64-bit systems.
Nowadays you can hardly find a developer who has a computer with a 32-bit microprocessor. But you still should remember that you must have a 64-bit computer that will let you manage all the tasks you have to solve. It seems that the most real case is when 64-bit hardware architectures still work under 32-bit operating systems. You should take into consideration expenses on purchasing and installing 64-bit versions of operating systems. Consider also additional expenses related to changing the version of the operating system, for example, reinstallation of various software.
Add the price of purchasing and mastering new compilers able to create 64-bit code to the total price.
You might need to purchase 64-bit versions of libraries and other components. Find out beforehand about the pricing policy of those companies whose components are used in your project. Sometimes 32-bit and 64-bit versions of components are sold separately. If you use open-source libraries that have no 64-bit configurations yet, be ready to spend much time on modifying them manually.
Take into consideration the time needed for your employees to study all the necessary information on 64-bit system development. You may also need to buy some additional tools such as, for example, PVS-Studio.
As you already know from the previous lessons, compiling a 64-bit configuration is only the beginning. In most cases, you will need to find and correct a lot of defects that will occur in 64-bit code. It is perhaps the most laborious yet most difficult to estimate part of the work. However we can advise you the following way relying on PVS-Studio static analyzer.
Well, you have several (tens, hundreds) Mbytes of source code ready for migration. There is no 64-bit configuration of the code yet. So, there are no files to be compiled in the 64-bit mode as well.
PVS-Studio provides you with a capability to detect 64-bit code issues even in 32-bit projects. It is this capability that will let you estimate the difficulty of migration BEFORE creating the 64-bit configuration of your project.
I would like the readers to note how the check of the code is performed in the 32-bit mode. You should understand that this check cannot be considered complete and even if you correct all the errors detected, you cannot be absolutely sure that the code will work in the 64-bit mode. Code of any serious application has such fragments:
#ifdef WIN64 ... #endif
Of course, this fragment will be skipped when testing the code in the 32-bit mode. Or, more exactly, while there is no 64-bit configuration yet, the application code might have no such a fragment.
Here is another important thing: it is natural that data types differ depending upon the project configuration. That is why the check in the 32-bit and 64-bit modes will nearly always lead to different results.
But how much different will they be? According to the results of the experiments carried out in our company, we have the following: the lists of diagnostic warnings generated by PVS-Studio analyzer when testing projects in the 32-bit and 64-bit modes coincide 95-97%. It means that not more than 5% of diagnostic messages differ.
These results were obtained in the following way. We took the code of real projects, checked it in the 32-bit mode and saved the list of diagnostic warnings. Then we checked the code of the same projects in the 64-bit mode and also saved the list of diagnostic warnings. After that we compared these lists and estimated how many percent of the diagnostic messages coincided. Since the whole procedure was performed in automatic mode, the number of projects that were analyzed was enough (more than 20 projects with the code size of several Mbytes each). So we may conclude that the figures (5% difference) can be trusted.
Of course, you should not hurry to fix all the potential errors detected in the 32-bit configuration of your project - it is better to wait for the 64-bit configuration. But you can easily estimate how much time you will need to check all the warnings generated by the code analyzer.
We recommend you to do the following:
- Analyze the 32-bit configuration of the project with PVS-Studio.
- One programmer who knows well the issues of 64-bit code looks through the warnings generated by the analyzer during a day and decides if this or that error is relevant to the project. If it is, the programmer corrects it.
- The total number of the analyzer-generated messages is divided by the number of the messages the programmer has looked through and processed during one day.
- The number you get is the number of man-days needed to port the application's code to a 64-bit platform.
The programmer must correct the errors found. It is not enough just to find an error and imagine that it is corrected. To detect and to correct are actions that differ in time they take. You may need to modify the program code in many project files to correct some errors. To avoid an understated estimate, you must make all the necessary corrections.
Of course, there is a drawback in this algorithm of estimating the migration process - it is the skill of the developer who will process the messages of the analyzer and modify the code during a day. So we recommend you to be very serious and careful when choosing a programmer responsible for the estimate.
Here are some recommendations on how to choose such a programmer:
- This person must be an experienced programmer who has been working in your company not less than for three years and who knows this particular project you want to port.
- The programmer must be familiar with the issues of 64-bit code - for example, know these lessons or the article "20 issues of porting C++ code on the 64-bit platform".
- It is desirable that the programmer understand the principles of working with static code analyzers. It is not an obligatory requirement but understanding the static code analysis technology makes the estimate of the migration process more adequate.
- The person must be able to stay within the usual working conditions during the testing day. He or she must not try to set up a record of performance to impress the colleagues. One cannot work all the days in such a way, and the terms will be estimated incorrectly.
Following these recommendations will allow you to get an adequate estimate of the cost and term of the 64-bit software migration process.
Consider the cost of adapting your testing system for full-fledged testing of 64-bit units. If your programs process large data amounts, you must have tests that run on data amounts more than 4 Gbytes. In its turn, integration of heavy tests might result in the task of test parallelizing. In this case you might have to buy additional tools.
If you use software or firmware systems of software copying and cracking protection, you should add the price of implementing protection for your 64-bit code to the total cost. Perhaps you will have to master new protection systems if those systems you are using at present do not support 64-bit codes. You might face other unexpected troubles, so make sure you have some time in reserve to manage them.
You will have to create a new distribution kit - this issue was considered in the previous lesson.
The rightholder of the course "Lessons on development of 64-bit C/C++ applications" is OOO "Program Verification Systems". The company develops software in the sphere of source program code analysis. The company's site: http://www.viva64.com.
Contacts: e-mail: email@example.com, Tula, 300027, PO box 1800.