An Analysis of Linux Scalability to Many Cores, OSDI 2010
Discussion lead: Ignacio Cano, Amrita Mazumdar, Liang Luo
What is a sloppy counter? What’s the key invariant for its correctness? Mention some of its benefits and limitations.
Describe one scalability problem in applications, its root cause, and the remedy.
We have seen two common approaches: to make the kernel implementation scalable (so that application developers don’t have to worry too much about it); or to expose low-level kernel abstractions so that applications could make better decisions on scalability. Which one (or some hybrid approach) do you think would extend to 96 or 128 cores, assuming that’s the hardware trend?
Do you think their benchmarks are realistic? Describe pros/cons and what would you add/remove.
Provide a list of questions you would like to discuss in class. Feel free to provide any comments on the paper and related topics (e.g., which parts you like and which parts you find confusing).
scalability: “A service is said to be scalable if when we increase the resources in a system, it results in increased performance in a manner proportional to resources added.”
Figure 1, accept - see