MOCA (Memory Organisation Cartography & Analysis) is a tool designed to provide a fine analysis of an application's memory usage. It mainly consist of a kernel module and a script which allows you to load the module and monitor a program.

MOCA is the successor if HeapInfo [1], a Valgrind tool which intercept every memory access and gives a carthograpy of the memory usage. This new tool is based on a Linux kernel module, it intercept pagefaults to record memory access. As pagefaults do not occurs very often, it periodically trigger false page faults to record more access. This mechanism is inspired by [2]. More details on Moca's implementation can be find on [3].

No tools is currently provided to visualize Moca traces, still there is a prototype of visualization using framesoc [4] and ocelotl [5]. The importer required to visualize Moca traces in framesoc will be available soon in framesoc importer repository.

To monitor a program with Moca, you can run the following command as root

# moca -c "my_command" -a "my_arguments"

It is also possible to retrieve data structure name addresses and size, this is done using a Pin instrumentation. To do so first install Pin from here. On 64bits systems you will also need to install 32 bits support, on debian it will be something like sudo apt-get install libc6-i386.

One this is done just tell Moca to use pin for structure detections:

# moca -c "my_command" -a "my_arguments"

At this point pin must be in root's path and either pin is installed in /opt/pin/ or PIN_HOME should point to pin installation directory.

The command output will be logged in the files Moca-cmd.log and Moca-cmd.err, these filenames can be modified with the option -f file.

The output of Moca is logged in the file Moca-output.log, this can be modified by the parameter -l file.

Finally the trace is in the file named Moca-full-trace.csv.

The following parameters allows you to do fine tuning on the module. By default you shouldn't need to use them.

• If you encounter performance issues, you can increase the wakeup interval, the priority (reduce the system noise) or the hashmap numbit parameter.
• If Moca tells you that a part of the trace have been dropped because there was not enought space to sotre it, you can increase the number of chunks, the chunksize or reduce the wakeup interval. Please note that, as memory is quite restricted in the kernel, it might be a better idea to play on the wakeup interval the priority than on the storage related parameters.
  • -w ms Set the wakeup interval for Moca.
    Default: 40ms
  • -p prio Schedtool priority for the kernel module, the program priority will be prio-1. You can increase this parameter to reduce the system noise.
    Default: use the normal process priority
  • -b numbits Set the number of bits used for the chunk hashmaps. The map size is 2^numbits. the higher numbits is, the less collision you should have.
    Default: 14
  • -S ChunkSize Set the number of adress that can be stored in one chunk. You can also reduce the wakeup interval, and therefore the number of adresses needed per chunks.
    Default: 2*2^14.
  • -C nbChunks Set the number of chunks kept in memory.
    Default: 20.
  • -F Disable false page faults. If you set this flag, almost all memory event will be missed, you really shouldn't use it ..."
  • -u Use ugly false page faults hack, this can reduce MOCA's overhead, however you must be sure that your application won't swap or it will crash !

Traces are saved as one human readable csv file. Each line indicate one memory accesse described by its virtual and physical addresses (@Virt and @Phy) the number of reads and writes (Nreads and Nwrites) a bitmask indicating on which CPUs the access have been detected (CPUMaask) a start and end timestamp correspondig to the chunk timestamp (̀Start and ̀End) and finaly the internal identifier of the task (thread) responsible for the access (TaskId).

MOCA output is designed to be imported inside the trace visualisation framework Framesoc[4], and to be visualised using Ocelotl[5]. This tool uses an adaptive algorithm to aggregate data. This algorithm returns a set of results which provides a trade-off between the information loss and the reduction of the visualisation complexity. The user has then the possibility to move the cursor between a very precise view trace or something more aggregated. Moreover it provides the ability to navigate through the trace, focus on one type of event or another.

To install Framesoc and ocelotl, go to the official website and follow the instruction. Than you can import Moca's trace through Framesoc's Moca importer and visualise it with Ocelotl.

**Note: ** This visualisation method is still experimental, the importer will soon be publicly accessible.

No installation script is provided yet, however moca script can be run from anywhere using the parameter -d to give the path to the Moca directory. If "-n" is not specified in the command, Moca will recompile the module.

## References

[1] David Beniamine. Cartographier la mémoire virtuelle d'une application de calcul scientifique. In ComPAS'2013 / RenPar'21 , Grenoble, France, 2013.
[2] E. H. M. Cruz, M. Diener, and P. O. A. Navaux. Using the Translation Lookaside Buffet to Map Threads in Parallel Applications Based on Shared Memory. In Parallel Distributed Processing Symposium (IPDPS), 2012 IEEE 26th International, page 532-543, May 2012.
[3] D. Beniamine, Y. Corre, D. Dosimont and G. Huard. Memory Organisation Cartography & Analysis. Technical Report RR-8694, Inria. March 2015.
[4] G. Pagano, D. Dosimont, G. Huard, V. Marangozova-Martin, and J. M. Vincent. Trace Management and Analysis for Embedded Systems. In Embedded Multicore Socs (MCSoC), 122, Sept 2013.
[5] Damien Dosimont, Lucas Mello Schnorr, Guillaume Huard, and Jean-Marc Vincent. A Trace Macroscopic Description based on Time Aggregation. Technical Report RR-8524, Apr 2014. Trace visualization; trace analysis; trace overview; time aggregation; parallel systems; embedded systems; information theory; scientific computation; multimedia appli- cation; debugging; optimization.