[panda-users] llvm-trace memory usage

[Brendan Dolan-Gavitt]

They are very large. However they are also highly compressible. One
way to reduce the space taken up is to modify panda/tubtf.c to use
gzip compression as it writes the file; I've done this in the past and
it works well, but it does mean that the resulting log file is not
seekable (i.e. it's hard to process it in any order other than the
beginning to the end).

We are working on a trace format that addresses these issues, but it's
not ready at the moment.

Best,
Brendan

On Wed, Feb 10, 2016 at 12:01 PM, mahboobe daftari
<mahboobehdaftari at gmail.com> wrote:
> Hi All,
>
> I am using llvm-trace plugin to replay a very short recording. The resulting
> TUBTF is 15 GB for the recording of almost nothing ( It is more than 40G
> when I run an echo server)! Is this the case all the time or am I doing
> something wrong? Even if not running TUBTF mode the llvm-memlog.log file is
> approximately the same size.
>
> I use the premade debian image and I did not add anything to that except for
> the echo server. I even tried disconnecting Internet, but the result was the
> same.
>
> Here is the way I replay the recording: "qemu-system-x86_64 -m 256 -replay
> test -panda llvm_trace:base=.,tubtf=1"
>
> Thank you in advance for the help,
> Mahboobeh
>
> _______________________________________________
> panda-users mailing list
> panda-users at mit.edu
> http://mailman.mit.edu/mailman/listinfo/panda-users
>



-- 
Brendan Dolan-Gavitt
Assistant Professor, Department of Computer Science and Engineering
NYU Polytechnic School of Engineering