Re: Advanced Internals - Madison

Well --- That's going to be hard to do. The spin lock is not the same as a mutex. The purpose of the spin lock is to protect against inter-process access to the same memory. The mutex is to protect against inter-thread access to the same memory. Every mutex has a spin latch associated with it to protect the mutex itself. There are literally thousands of spin latches that are dynamically created within the server to do things like protect link lists, protect the memory queues, etc. For instance, when we are updating somthing in memory, then it's probably going to have a spin latch associated with it. Replicated transactions have multiple spin latches per transaction. Another thing about spin latches... One of the things that a spin latch does is to flush the memory in the cpu cache back to main memory. This is an important thing to consider when you think about a 64-bit long long on a 32-bit machine. Obviously, there is no way that a 32-bit machine can ensure that a 64-bit integer is flushed back to main memory as an atomic unit. So the memory flush of the dirty CPU cache memory is coordinated with the spin latch. <david@smooth1.co.uk> wrote in message news:1115856009.286457.132360@f14g2000cwb.googlegroups.com... > > New FAQ entry for Madison > > "On 5th June 2001 mpruet@home.com (Madison Pruet) wrote:- > > The output of onstat -g spi is only a few of the more critical > spinlocks." > > > David asks so how do we see the rest? So undocument stuff > available so we can see them? >