Description:
Modern computer systems are called on to deal with billions of events every second, whether they are instructions executed, memory locations accessed, or packets forwarded. Sifting through these data streams has many applications, from helping to identify attackers (e.g. intrusion detection), enabling higher performance (profiling), or assisting in debugging and reliability (introspection). As our applications and software continue to grow in complexity it is worth considering a redesigned computing device that supports online high-throughput deep event analysis as a first class citizen -- everywhere from the circuits to the software. In particular I will talk about two complementary techniques. 3d-introspection (a circuit/architecture method) offers the potential of modular ``snap-on'' functionality that could be included with developer systems, and omitted from consumer systems to keep the cost impact to a minimum. We show that hardware stubs could be inserted into to commodity processors at design time that would allow analysis layers to bonded to development chips, and that these stubs have surprisingly low area and power overheads. Our second idea, Range Adaptive Profiling, is a new profiling method capable of automatically extracting hierarchy from streams of data (application, system, or hardware events) with unknown structure. In this talk, I will describe the concepts behind these idea, describe how they can be performed efficiently in software and hardware, and outline our recent prototyping work and extensions.
Speaker(s):
Tim Sherwood, assistant professor, Computer Science, UC Santa Barbara
|
