From Wikipedia, the free encyclopedia
|WikiProject Computer science|
There's several things about this article which strike me as a bit iffy, but I'm not a computer scientist so I don't want to start chopping away.
- The references to "idle wait" in kernel-level critical sections. I don't know of any architecture that does this; spinlocks ie. "busy waits" are standard in kernel space.
- "Typical values for L/E range from 0.01 to 0.1." The reference for this statement is from 1968! By a conservative estimate, computers have gotten at least a billion times faster since then.
- The article treats the hardware itself as being contention-free, suggesting an overly theoretical bias; in practice hardware-level contention has informed the development of Linux and presumably other multi-threaded kernels.
- "Software lockout is the major cause of scalability degradation in a multiprocessor system"--if this were true we could just run a distinct kernel image on each processor of a NUMA system, consider the problem solved, and take the rest of the day off to play golf.
--188.8.131.52 23:23, 4 September 2007 (UTC)
- Spinlocks are often used in kernel space, but shouldn't be used unless the typical time for the wait is less than, say, the time for a context switch; I think at least some systems have "adaptive" locks that spin for a little while and then block to allow the CPU to go do something else, if it looks as if the lock won't get released for a while. In any case, it's a "busy wait" only in the sense that the CPU is busy spinning its wheels; it's not as if the CPU is doing useful work while spinning. Perhaps it should just refer to "wait times". Guy Harris (talk) 01:47, 5 November 2011 (UTC)