Presentation is loading. Please wait.

Presentation is loading. Please wait.

02 de Octubre de 2016 QNX 6.4 - QNX 6.3 COMPARISON.

Similar presentations


Presentation on theme: "02 de Octubre de 2016 QNX 6.4 - QNX 6.3 COMPARISON."— Presentation transcript:

1 02 de Octubre de 2016 QNX 6.4 - QNX 6.3 COMPARISON

2 QNX 6.4 - QNX 6.3 Comparison 2 The Workbench

3 QNX 6.4 - QNX 6.3 Comparison 3 THE WORKBENCH Note: Each machine have two disk partitions (QNX 6.4.0 / QNX 6.3.0+Core Patch 6.3.2A both standard installation)

4 QNX 6.4 - QNX 6.3 Comparison 4 Networking Throughput - Message Passing

5 QNX 6.4 - QNX 6.3 Comparison 5 Networking throughput - Message passing To compare efficiency of message passing over QNET network Goal A variable amount of messages is sent between two nodes … for example: 10 times, 5000 messages The elapsed time is measured Two diferent times are taken: the average time and the total time … for example the average time to send 5000 messages and the total time to send 10 times 5000 messages (50000 messages) This operation is repeated under QNX 6.3 (io-net) and QNX 6.4 (io-pkt) The results are compared Method

6 QNX 6.4 - QNX 6.3 Comparison 6 RESULTS TABLE A smaller efficiency of aprox 2,62% is observed in QNX 6.4 (Why 1?) Side Note: Using io-pkt-v4 manager the smaller efficiency is still more evident (-5.6%)

7 QNX 6.4 - QNX 6.3 Comparison 7 TOTAL TIME GRAPH

8 QNX 6.4 - QNX 6.3 Comparison 8 AVERAGE TIME GRAPH

9 QNX 6.4 - QNX 6.3 Comparison 9 File-system Throughput

10 QNX 6.4 - QNX 6.3 Comparison 10 FILE-SYSTEM THROUGHPUT To compare efficiency of writing/reading to the filesystem (disk) Goal A simple process opens, writes and reads a disk file (byte by byte) … first calling open, write, read. … then calling fopen, fwrite, fread. The elapsed time is measured Two different times are taken: in fd way and stream (FILE*) way The results are compared This operation is repeated under nodes DEV1 and DEV2 Method

11 QNX 6.4 - QNX 6.3 Comparison 11 DEV1 RESULTS TABLE

12 QNX 6.4 - QNX 6.3 Comparison 12 DEV2 RESULTS TABLE

13 QNX 6.4 - QNX 6.3 Comparison 13 DEV1 GRAPHS Conclusions (DEV 1) Stream method (FILE*) is much more efficient than file descriptor method QNX 6.3 is much more efficient in file descriptor method than QNX 6.4 QNX 6.3 is still more efficient in stream (FILE*) method than QNX 6.4

14 QNX 6.4 - QNX 6.3 Comparison 14 DEV2 GRAPHS Conclusions (DEV 2) Stream method (FILE*) is much more efficient than file descriptor method QNX 6.3 is much more efficient in file descriptor method than QNX 6.4 QNX 6.3 is still more efficient in stream (FILE*) method than QNX 6.4

15 QNX 6.4 - QNX 6.3 Comparison 15 SOME NOTES The difference between fd and FILE* efficiency is at least understandable. Reading and writing on a file descriptor is almost a direct system call (with much more background work), whereas reading and writing on FILE* is a wrap of file access for which the stdio API take place. This “background work” (via fd) is harder in QNX 6.4 than in QNX 6.3 (Why 2?) Even working with streams (FILE*) QNX 6.3 seems to be more efficient (Why 3?) (Is it correct ?)

16 QNX 6.4 - QNX 6.3 Comparison 16 RESULTS TABLE FOR LARGER FILES Note to avoid confusion: 400000 KB = 400000 x 1024 bytes = 409600000 bytes = 390.625 Mbytes (test unit was KB)

17 QNX 6.4 - QNX 6.3 Comparison 17 SOME NOTES To W/R 400000 KB in DEV1 it takes 225.50 secs in 6.3 while in 6.4 it takes 227.54 secs To W/R 400000 KB in DEV2 it takes 198.89 secs in 6.3 while in 6.4 it takes 200.76 secs The average time is similar in DEV1 and 1.46% faster in QNX 6.3 for DEV2 (Why4?) DEV2 is 12.39% faster than DEV1 in 6.4 and 14.34% in 6.3 (Why 5?) Because DEV1 (2.8 GHz – 2MB RAM) is dual core with more RAM than DEV2 (3.2 GHz – 1MB RAM)

18 QNX 6.4 - QNX 6.3 Comparison 18 VERY BIG FILE TEST Writing and then reading byte by byte 2000000 KB stream (FILE*) takes in DEV1: 1149.7 [segs] in QNX 6.4 1139.3 [segs] in QNX 6.3 Difference: 10.4 secs faster in QNX 6.3 (  0.9%) (Why 6?)


Download ppt "02 de Octubre de 2016 QNX 6.4 - QNX 6.3 COMPARISON."

Similar presentations


Ads by Google