Re: [RFC][PATCH] ipc: Remove IPCMNI

Manfred Spraul [off-list ref] writes:

Hello Mathew,

On 03/29/2018 12:56 PM, Matthew Wilcox wrote:

quoted

On Thu, Mar 29, 2018 at 10:47:45AM +0200, Manfred Spraul wrote:

quoted

quoted

quoted

quoted

quoted

quoted

This can be implemented trivially with the current code
using idr_alloc_cyclic.

Is there a performance impact?
Right now, the idr tree is only large if there are lots of objects.
What happens if we have only 1 object, with id=INT_MAX-1?

The radix tree uses a branching factor of 64 entries (6 bits) per level.
The maximum ID is 31 bits (positive signed 32-bit integer).  So the
worst case for a single object is 6 pointer dereferences to find the
object anywhere in the range (INT_MAX/2 - INT_MAX].  That will read 12
cachelines.  If we were to constrain ourselves to a maximum of INT_MAX/2
(30 bits), we'd reduce that to 5 pointer dereferences and 10 cachelines.

I'm concerned about the up to 6 branches.
But this is just guessing, we need a test with a realistic workload.

Yes.

My primary purpose with the patch was to show that the issues with the
current limits could be resolved in a straght forward manner.  I really
don't know if idrs are the appropriate data structure.  It is possible
rbtrees are a better fit.


I think my algorithm I proposed for generating identifiers is as likely
as any to be a good one.  It does need testing on a wide variety of
applications to see what applications actually care about and for that I
think my proposed patch is more than sufficient.

By not keeping a generation counters in the slots themselves linux
already differs substantially from traditional implementations.

Doing something to free us from using a fixed number of bits for the
counter and a fixed number of bits to encode the slot we can support
much larger use of this API.

Eric