11/5 two Reviews

[Juexin Wang]

Title:
BorderPatrol: Isolating Events for Black-box Tracing

Authors:
Eric Koskinen,  John Jannotti 

Date:
April. 2008

Novel Idea:
Present a tracing technique that actively isolates black-box inputs so that request paths can be precisely observed, without materially affecting the overall application’s ability to multiplex requests.

Main Result(s):
Implement BorderPatrol, which obtains precise request traces through systems built from a litany of unmodified modules.   
BorderPatrol obtains precise traces for black-box systems  that cannot be traced by any other technique.

Impact:
Provide a precise way to obtain the protocol-specific trace in black-box systems without specific application instrumentation.

Evidence:
- BorderPatrol obtains traces using active observation which carefully modifies the event stream observed by modules, simplifying precise observation. 

- Protocol processors leverage knowledge about standard protocols, avoiding application-specific instrumentation.

- Event isolation unbundles concurrent input events in order  to allow the observation of a module’s behavior on a perevent basis. 

- Message witnesses identify matched messages, usually request/response pairs. 

- Event isolation and message  witnesses are provided by protocol processors,  an abstraction that allows developers to implement protocol-  specific tracing.

- BorderPatrol seeks to follow the repeated transfer of a  request from one black-box module to another in order to construct causal paths that show which modules handled a given request, in what order, and for how long.

- Request traces can be thought of as chains that are made up of two types of links: external and internal links.

- BorderPatrol makes inferences based on the  assumption that the internals of black boxes are honest, immediate, and independent.

-In active observation, it employs protocol processors, message witnesses, event isolation.

Prior Work
Some other works of Instrumentation, pervasive frameworks, probabilistic correlation and causal paths.

Competitive work
None, BorderPatrol first show us that traces can be obtained for unmodified programs without sacrificing precision.

Reproducibility
N/A

Question/Criticism:

N/A

================================================================================

================================================================================

Title:

 Performance debugging for distributed systems of black boxes

Author(s)

AGUILERA Marcos K.; MOGUL Jeffrey C.; WIENER Janet L. ; REYNOLDS Patrick

Date:
Oct. 2003

Novel Idea:
Used two methodologies, nesting in which RPC calls were matched and the other convolution in which they tried to match causal calls by simply doing statistical analysis.

Main Result(s):
Implement tools in two ways that enable programmers to isolate performance bottlenecks in distributed systems composed of black-box nodes. 

Impact:
Approach this problem by obtaining message-level traces of  system activity, as passively as possible and without any knowledge  of node internals or message semantics. The algorithms they designed can ascribe delay to specific nodes on specific causal paths. requires no  modifications to applications, middleware, or messages. 

Evidence:

- High latency paths and High latency access patterns.

- Nesting: Looks at individual paths and aggregates, find rare paths. Faster, more accurate, limited to RPC-style systems;

- Convolution: works for all message-based systems. need set a appropriate time step.  

- Identify causal path: causality trace by timestamp.

- make trace: synthetic trace generator.

Prior Work
Not mentioned

Competitive work
In section 3.1, some works aiming on similar problems but take less radical  approaches to the problem of black-box components and require either a homogeneous implementation environment or more intrusive instrumentation.

Reproducibility
N/A

Question/Criticism:

Nesting algorithm may need to store a lot of information in memory. 

-- 
J.W
Happy to receive ur message