-
Notifications
You must be signed in to change notification settings - Fork 0
Phone Conference, November 17th 2011
Benchmarking and non-regression tests are elements which can become critical in the process of developing a file server. This is not an easy subject.
Many OpenSource tests already exist, like the well-known iozone, bonnie++ or IOR. Very often, benchmarks are used to stress the file server in order to produce bugs and identify them. Some tests are private to a company or an institution and designed to check a specific part of the code. Sometimes, this job is part of a dedicated teams, separated from the developers' one. Anyway, the trend is to use benchmark with configurable workload (SpecSFS, dbench, filebench), having fewer benchmarking utilities but more profiles.
It was hard to define a clear trend for all contributors, and the discussion should follow on the list. A wiki page has been created to collect this information. The idea is to share common practices in order to be able to build an exhaustive and useful test suite to check non-regression and performances on Ganesha.
A bug exists with NFSv4.0 locks. It's located in the NFSv4 implementation and is probably there for a long long time. Philippe will send elements on this on the mailing list (including wireshark captures)
There is a need for defining a state diagram which will describe how to manage a file holding state and been invalidated. A document (probably under the shape of a "picture" showing a diagram) is to be made. Its content will be discussed on the list, including various scenarios.
Frank and JV discovered a bug in the permission checkings done when trying to acquire a lock under NFSv3/NLMv4. An issue has been created for this (http://github.com/phdeniel/nfs-ganesha/issues/24)
No pending subjects for the moment. Because of Thanksgiving, next week conference is canceled. Next conference will be done in two weeks.