LogP: Towards a Realistic Model of Parallel Computation (1993) [pdf]

people.eecs.berkeley.edu

12 points·espeed··3 comments

LogP was based mostly on experiences in analyzing and working with parallel machines in the 1990s, best summarized in Culler, Singh, and Gupta's book "Parallel Computer Architecture: A Hardware/Software Approach". The model was based upon lessons learned from trying to build fast message layers for those machines. The influence was mostly aimed at the software layers of those machines, which led to Active Messages, U-Net, and related low-overhead messaging layers; less influential upon architectures, I would think.

Seems similar to the Cell architecture of PS3 (if you look at how all the "Synergistic Processor Elements" are networked together via a bus). I wonder whether it was based on this?

LogP is a general-purpose adaptive parallel model so not tied to a specific architecture. However, I too am curious if anyone knows the Cell lineage and if there is a connection. A cursory search on Google Scholar [1] shows the LogP paper is cited more than 2000 times, but I did not see a direct lineage reference between the IBM Cell architecture [2] and the Berkeley LogP model [3]. Does anyone know the history and if there is a connection?

[1] https://scholar.google.com/scholar?cluster=52811921455145789...

[2] https://en.wikipedia.org/wiki/Cellular_architecture

[3] https://en.wikipedia.org/wiki/LogP_machine

Reply on news.ycombinator.com