Review: BorderPatrol: isolating events for black-box tracing
0 views
Skip to first unread message
Sunil Mallya
Nov 4, 2009, 11:58:55 PM11/4/09
to CSCI2950-u Fall 09 - Brown
Paper Title
BorderPatrol: isolating events for black-box tracing
Author(s)
Koskinen, Eric and Jannotti, John
Date
ACM SIGOPS/EuroSys European Conference on Computer Systems 2008
Novel Idea
A tracing tool to obtain traces without modifying components of a
system, but at the same time maintain precision.
Summary & Main Result(s)
The idea of this paper is to develop a technique to obtain traces
without modifying the system that is obtain traces in a system made up
of black boxes. The authors start by defining what “Black-box” in this
context means and how oblivious their technique is of the internal
modules, the assumptions are clearly stated and explained in detail.
The main technique introduced in the paper is active observation in
which the tool observes and subtly modifies the event streams sent and
received by monitored modules. The tool itself cannot be deployed for
a unknown system, but it has the knowledge of what it is monitoring to
the extent of logging application specific identifiers but without the
knowledge of the internals of that module.
There is an important section in the paper where the authors evaluate
their assumptions against a real world scenario which arrives at the
conclusion that BorderPatrol works as the tool is able to ensure that
events are serialized even in multi threaded components. Also it is
able to identify request boundries and contain no application
specifics. Finally results indicate that the tool allows safe
employment mechanism that isolates inputs events arriving at a
component without affecting its capability to be multiplexed.
Impact
Traces are useful to the developers to analyze and detect performance
related issues and also they can be valuable inputs to automated
systems. BorderPatrol could be helpful in putting together large scale
systems and tuning them for performance enhancements or even
performance checks.
Evidence
The main objective of evaluation here is divided into two parts –
Prove that BorderPatrol can trace systems with black-boxed components
& secondly evaluate itself with previous systems which required manual
instrumentation. They evaluate the system on Muti-threaded tier
(dearinter.net) and Zeus (event-driven Webserver).
I am convinced with their evaluation and it’s good that the paper
gives us in good detail their evaluation and more importantly they
show that tracing in black-boxed system can be done efficiently.
Reproducibility
Yes, BorderPartrol binaries are available on the website and enough
information is given for the results to be reproduced.
Criticism
I am not sure about if the overhead on the system is too much (10-15
%) while running BorderPatrol. As I mentioned earlier this overhead
may be ignored when BorderPatrol is used for tuning the system for
better performance.
BorderPatrol: isolating events for black-box tracing
Author(s)
Koskinen, Eric and Jannotti, John
Date
ACM SIGOPS/EuroSys European Conference on Computer Systems 2008
Novel Idea
A tracing tool to obtain traces without modifying components of a
system, but at the same time maintain precision.
Summary & Main Result(s)
The idea of this paper is to develop a technique to obtain traces
without modifying the system that is obtain traces in a system made up
of black boxes. The authors start by defining what “Black-box” in this
context means and how oblivious their technique is of the internal
modules, the assumptions are clearly stated and explained in detail.
The main technique introduced in the paper is active observation in
which the tool observes and subtly modifies the event streams sent and
received by monitored modules. The tool itself cannot be deployed for
a unknown system, but it has the knowledge of what it is monitoring to
the extent of logging application specific identifiers but without the
knowledge of the internals of that module.
There is an important section in the paper where the authors evaluate
their assumptions against a real world scenario which arrives at the
conclusion that BorderPatrol works as the tool is able to ensure that
events are serialized even in multi threaded components. Also it is
able to identify request boundries and contain no application
specifics. Finally results indicate that the tool allows safe
employment mechanism that isolates inputs events arriving at a
component without affecting its capability to be multiplexed.
Impact
Traces are useful to the developers to analyze and detect performance
related issues and also they can be valuable inputs to automated
systems. BorderPatrol could be helpful in putting together large scale
systems and tuning them for performance enhancements or even
performance checks.
Evidence
The main objective of evaluation here is divided into two parts –
Prove that BorderPatrol can trace systems with black-boxed components
& secondly evaluate itself with previous systems which required manual
instrumentation. They evaluate the system on Muti-threaded tier
(dearinter.net) and Zeus (event-driven Webserver).
I am convinced with their evaluation and it’s good that the paper
gives us in good detail their evaluation and more importantly they
show that tracing in black-boxed system can be done efficiently.
Reproducibility
Yes, BorderPartrol binaries are available on the website and enough
information is given for the results to be reproduced.
Criticism
I am not sure about if the overhead on the system is too much (10-15
%) while running BorderPatrol. As I mentioned earlier this overhead
may be ignored when BorderPatrol is used for tuning the system for
better performance.
Reply all
Reply to author
Forward
0 new messages