Edward, I am not sure if you had a chance to read my last reply. Thinking more about this I wanted to ask you a few questions.
There could be a way that more than one person is involved. For the current accrual algorithm we did not have to build many of the parts from total scratch what do you anticipate is the overall lift, looking over the gossip code does swim fit in well or do you think there is a big retro-fit required. I ask because you seem to like/understand the swim protocol. Thanks, Edward On Fri, Mar 10, 2017 at 6:44 PM, Edward Capriolo <[email protected]> wrote: > > > On Fri, Mar 10, 2017 at 6:17 PM, Edward McEnrue <[email protected]> > wrote: > >> Hi Everyone, >> >> Thanks so much for the GSOC opportunity! I'm a potential candidate that is >> interested in applying to work on implementing GOSSIP-51 (SWIM, a >> beautiful >> protocol) as well as any other tasks if time permits. >> >> I have some questions about the current implementation: >> >> 1: are tests run with randomized packet loss? >> 2: what is the current underlying implementation of gossip based on? That >> is, is there an academic paper that I could read to help me understand the >> code base further, so that I may delineate what API changes must change? >> 3: Have there been prior discussions regarding implementing SWIM that I >> could use for my proposal? >> >> Thank you, >> >> Edward McEnrue >> > > Edward, > > Hello and welcome. You have the right name to make a big impact around > here! JK. > > The situation is this: We already have one GSOC applicant and the swim > ticket is earmarked for them. I do not know how many applicants /proposals > are assigned per project, but you are the 4th person to express an interest > in Gossip this week. It would be great if you look through Jira and find > some other issues inside the project you are interested in and we can add > the GSOC label to those. > > I will answer your questions and make suggestions. > > 1) are tests run with randomized packet loss? > No. We have an integration type test that terminates a node mid test > asserts convergence restarts the node and asserts convergence. > https://github.com/apache/incubator-gossip/blob/ > bcd10510ee7da41322bf2aece4b0ef7e4789f199/src/test/java/org/apache/gossip/ > ShutdownDeadtimeTest.java > We also have some other tests that test the math in the accrual detector > directly > https://github.com/apache/incubator-gossip/blob/ > bcd10510ee7da41322bf2aece4b0ef7e4789f199/src/test/java/org/ > apache/gossip/accrual/FailureDetectorTest.java > > There are ways we could randomly inject failures into the process as you > have described. That would not be a bad ticket if you are interested. > > 2) what is the current underlying implementation of gossip based on? > We are doing something similar to this: http://fubica.lsd.ufcg.edu. > br/hp/cursos/cfsc/papers/hayashibara04theaccrual.pdf > > 3) Have there been prior discussions regarding implementing SWIM that I > could use for my proposal? > No. But as I said before it would probably be better to look for something > other than SWIM. > > You are also welcome to take on some tickets not marked GSOC and get > involved or propose some other features/improvements. > > Thanks, > Edward >
