Computer Systemen 2013: Labbook

16 October, 2013

Telnet was missing, so copied one to /opt/prac/cs_ai/packages/valgrind/bin.
Checking the hardware of the crazyfly.
The Crazyfly is running on-board an OpenOCD server, which listens to telnet- (port 4444) and gdb- (default 3333) clients.
According to the firmware code, the stm32 processor is running FreeRTOS.
Inside FreeRTOS there is a webserver example:

Copied cpuid from /home/arnoud/bin/linux (on mremote) to /opt/prac/cs_ai/bin. The cpuid version is 2010324. Could look for a more recent version (reports that u019948 is a Xeon).
In /home/arnoud/src/cpuid-20060917 the source code of this program is available. The program was written by Tod Allen from etallen.com.
A more recent version of cpuid could be found at Allen's cpuid page. I downloaded version 20130610. Installed this version on /opt/prac/cs_ai/bin. Correctly identifies u019948 as Core-i3 processor. Still, the cache information is confusing.
Command more /proc/cpuinfo didn't give much more information.
Looked up the information for a i3-2120 processor. Double core, four virtual cores, clockspeed 3.3GHz. No cache size given. Found an indication at cpu-world: L1 (2x 32 Kb instruction and data caches), L2 2x256 KB, L3 3Mb.
Also looked up acheron (E5420). Xeon processor, 4 cores, 4 virtual cores, 2.5 GHz. According to cpu-world: L1 (4x32 KB data and instruction cache), L2 2x 6Mb, no L3).

Tried to print in G0.18. As suggested, I did //uvaprint.uva.nl on Start->Run of nb-udk. With my UvAID@uva.nl I could see the printers, but P005435 was not with them. Could not find my labbook of the Linux workstations, where I tried something comparible from the machines in the room. Copied the file to acheron, and printed with lp -d P005435 TentamenSep2012.pdf. The pdf is missing question 4e-g.
Testing cachelab with ./test-trans -M 32 -N 32 fails on csim.c. The reference transpose function has hits:870, misses:1183, evictions:1151 which gives a score 0:2147483647. When I do ./driver.py gives no points because the transpose from pranav the misses are invalid.
Start with a simple copy. This also fails, but this is due that I didn't compile. The simple copy gives 0 point for part B. The reference gives 26 points for part B (287 misses ipv 1183, 1107 ipv 4723, 1913 ipv 4423). Checked the reference csim.c. Quite complete code, should look at a bit more simple implementation to distribute. Otherwise I should distribute the object code only.

Tried to create a writeup for cachelab, but my diskquota are full. Copied my scripts from mremote, but huge.sh has troubles with the spaces in the group. Solved this by using ls -o instead of ls -l.
overflow.sh has problems with circular definitions in the .snapshot directories. Found this explaination about this hidden directories. Yet, script seems to work (result in /tmp/huge.

Read Cachelab writeup. The lab consists of two parts: I don't know if I find the first part that importent for AI-students so first concentrate on part B (optimize transpose). The writeup suggest that solution has to be found in implementing a blocking algorithm.
Part B requires the program valgrind, which seems not to be automatically installed. Checked developer site, but no RPMs are available. Should be installed from the source code.
Did a ./configure --prefix=/opt/prac/cs_ai/packages/valgrind-3.8.1, which gave me the summary Primary -DVGPV string: -DVGPV_amd64_linux_vanilla=1.
Install failed because I am out of quota. Moved valgrind-source to scratch and tried again. make install fails in the directory VEX. Continued with doing make install in the other directories. The suggested test in the README seems to work: ./valgrind ls -l:
==18984== HEAP SUMMARY: ==18984== in use at exit: 19,689 bytes in 15 blocks ==18984== total heap usage: 224 allocs, 209 frees, 79,713 bytes allocated ==18984== ==18984== LEAK SUMMARY: ==18984== definitely lost: 0 bytes in 0 blocks ==18984== indirectly lost: 0 bytes in 0 blocks ==18984== possibly lost: 0 bytes in 0 blocks ==18984== still reachable: 19,689 bytes in 15 blocks ==18984== suppressed: 0 bytes in 0 blocks
Modified my bashrc and created a softpkg file for valgrind-3.8.1. Tested with ./test-trans -M 32 -N 32 and the test worked!

There seems to be not much progress with the implementation of C99 features in GCC. Instead, most effort seems to be dedicated to implement features of the 2011 standard.

/opt/prac/cs_ai. Made 22 puzzles from phases with command ./makepuzzle.pl -l Fall2013 -i 22 -u group22 -s ../src -o /opt/prac/cs_ai/datalab/2013/handout/ (which creates directory group22 in that directory).

Checked the labs. The datalab was corrected Sept 9 2012, 6 days after my last assignment. Also the bomblab was assigned later (Spring 2013).
Only difference in datalab is line 147 in btest.c. Difference in bomblab is in makebomb.pl script.
Copied the btest.[ch] into ./src and ./puzzles. Made a puzzle with command ./makepuzzle.pl -l Fall2012 -i 1 -u group1 -s ../src/ -o ../datalab-handout (after deleting the old puzzles).

Try to add my display at nb-ros. I see the second display of G0.18. The external display has many resolution options, my native display only one (which probably makes it impossible to mirror the displays). The display is to the right. I can drag documents to that location, but easily lose my mouse there.
To add resolution I seem to have to use xrandr. askubuntu gives also some clues about for instance lshw.
lshw indicated that nb-ros use Ivy Bridge graphics with the i915 driver. At the driver documantation found a page dedicated to adding undetected resolutions. First checked the /var/log/Xorg.0.log; /usr/lib/xorg/modules/drivers/intel_drv.so is loaded (version 2.17.0). eDP1 is probed for alternative modes, but only the same resolution (but lower frequency is reported).
Intel driver seems already to be the most recent. Found on Ubuntu wiki a link to five Graphical XRandR front ends. According to LXRandR ARandR is to most powerfull, yet I didn't see any options to add resolutions.
Checked Intel site. Our chipset is Intel® HD Graphics 4000, reported by Xconf as Ivybridge Mobile (GT2). Intel redirected to 01.org, but the Linux Graphics Installer was already obsolete (Ubuntu 13.04). Downgrade script updated 10 drivers (relevant for intel-graphics).