Vector VN8970 User Manual download pdf (Page 28)

Chapter 4. Performance

data extraction were run in order to have a reference and also to evaluate what dif-

ference the ioDebug part - the part of the software that sends the debug/test speciﬁc

frames over CAN, see Chapter 2.5 - of the software makes to the overall perfor-

mance. In order to decide how large amount of data to extract during the tests the

ioDebug code was studied to actually see what variables are sent on the bus. In the

case of the project used during testing ioDebug sent variables adding up to approx-

imately 1 kB of data. Normally during a test not all variables are necessary which

is why separate tests with different amount of data extracted were performed - one

with all the data, one with approximately half and one with approximately a tenth

of the data. Furthermore, in order to stress the system to its max, one test where the

whole RAM was extracted was performed. All of the tests with simultaneous data

extraction were done on an ECU with the Application software ﬂashed.

The performance script presents the result from the tests by means of execution

time and CPU-load for all the different so called schedules in the CPU - initiali-

sation schedule, operating schedule, suspended schedule, shut down schedule and

so on, which basically are different modes in which the CPU can be operating. It

is implemented so that it presents the worst case times and loads recorded for the

different schedules and threads. That is, the result presented below shows the max-

imum execution time of that thread recorded during the test. The most important

schedule to investigate is the operating schedule since it is in this schedule the CPU

is operating when everything is normal.

Execution times and loads might change slightly due to different noise factors

which is why each test case was run three times to get statistical validity. The result

is then presented as a mean value of the three measurements.

4.2 Result

Figures 4.2 and 4.3 show the accumulated time for all the threads in the different

schedules for the different software and amount of data extracted. Even though the

differences are small in absolute time (Monitor is roughly 2% slower than Applica-

tion in operating mode) it is clear that the Application build is the fastest and that

when the whole RAM is extracted is the slowest. When extracting the amount of

data corresponding to 10% of the ioDebug variables no noticeable difference can be

seen.

More importantly, the more realistic cases when data corresponding to all, or a

part, of the variables in ioDebug are extracted are not worse than the Monitor build.

All of the schedules (not presented here) show a similar behaviour which is why it

is possible to conclude that, for these purposes, the JTAG data extraction method

satisﬁes the current needs of not degrading the CPU’s performance.

1 2 ... 23 24 25 26 27 28 29 30 31 32 33 ... 66 67

No comments

Vector VN8970 User Manual Page 28