Re: VM object cache misses
Re: VM object cache misses
- Subject: Re: VM object cache misses
- From: Michael Smith <email@hidden>
- Date: Fri, 15 Jan 2010 08:30:30 -0800
On Jan 15, 2010, at 1:40 AM, Quinn wrote: At 15:16 +1100 15/1/10, Godfrey van der Linden wrote: and why is its hit-rate so low? I don't understand the dynamic behaviour of the VM subsystem well enough to answer that question. I will note that the cache is pretty small (vm_object_cached_max defaults to 512).
All this means is that your workload isn't characterised by the sort of behaviour the cache is aimed at. There are several behaviours like this in the Darwin kernel, where objects (vm_object_t, vnode_t, physical pages) are kept around in a valid but disassociated state after their explicit connection to a user process is torn down by that process.
This is because there are a number of not uncommon pathological workloads where these objects are set up, used and torn down many times over by one or more user processes, and the cost of creating these objects from scratch can be nontrivial.
Consider for example the traditional compiler workflow, where each stage in the preprocess/compile/assemble/link chain writes an intermediate file that is in turn read by the next stage.
= Mike
-- Excellence in any department can be attained only by the labor of a lifetime; it is not to be purchased at a lesser price -- Samuel Johnson
|
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Darwin-kernel mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden