About :: TODO :: Blog :: RSS :: Old blog :: Projects :: GIT :: Gallery :: Notes

Fri, 03 Nov 2006

Kevent roadmap.


I plan to implement following features in kevent in order as they are presented:

• new ring buffer implementation.
  It will be exactly like existing one, except that it will work with provided by userspace using special syscall.
  This will allow to make all kevent syscalls as so called 'cancellation points' by glibc, i.e. when thread has been cancelled in kevent syscall, thread can safely removed and no events will be lost, since each syscall will copy event into special ring buffer, accessible from other threads or even processes (if shared memory is used).
  Actually it is already possible, and it was there from the beginning of the kevent - one can provide shared memory (or memory used as shared buffer between threads) to kevent_get_events(), but in this case userspace should manage indexes (offsets where to put new ready events) by itself.
• wake-up-one-thread flag. When several threads wait on the same kevent queue and requested the same event, for example 'wake me up when new client has connected, so I could call accept()', then all threads will be awakened when new client has connected, but only one of them can process the data. This problem is known as 'thundering nerd problem'. With new flag it will be possible to mark some events, so only threads requested to be in the thundering nerd (i.e. those threads which have not set the flag on given event) would be awakened when event is ready.
• FIFO read/write notifications.
• userspace notifications. Userspace will be able to register a special event, which can be marked as ready by some other thread. In theory it is just usual pipe between users, so it is possible (and likely) that I will not do it.
• signal notifications. Unlikely to be implemented by me - I already have huge experience in 'working' with signal/rt-signal gurus, so additional project will likely mar my karma even more.

/devel/kevent :: Link / Comments ()