Indiana University-Purdue University Indianapolis (IUPUI) / This thesis presents a novel technique and framework for decreasing instrumenta-
tion overhead in software systems that utilize dynamic binary instrumentation. First,
we introduce a lightweight networking framework combined with an easily extensible
BSON implementation as a heavy analysis routine replacement. Secondly, we bind
instrumentation and analysis threads to non-overlapping cpu cores--allowing analysis
threads to execute faster. Lastly, we utilize a lock-free buffering system to bridge the
gap between instrumentation and analysis threads, and minimize the overhead to the
instrumentation threads. Using this combination, we managed to write a dynamic
binary instrumentation tool (DBI) in Pin using Pin++ that is almost 1100 % faster
than its counterpart DBI tool with no buffering, and less than 500% slower than a
similar tool with no analysis routine.
| Identifer | oai:union.ndltd.org:IUPUI/oai:scholarworks.iupui.edu:1805/10906 |
| Date | January 2016 |
| Creators | Du, Mike |
| Contributors | Hill, James Haswell, Tuceryan, Mihran, Raje, Rajeev R., Fang, Shiaofen |
| Source Sets | Indiana University-Purdue University Indianapolis |
| Language | en_US |
| Detected Language | English |
| Type | Thesis |
Page generated in 0.1029 seconds