Zane, Forwarding to list - I think you hit reply to instead of reply all :)
Regards -steve On 03/10/2011 07:42 PM, Zane Bitter wrote: > > On 2011/03/10, at 03:04, Steven Dake wrote: > >> This patch looks pretty good. >> >> One change that needs to be made before this patch can be merged is that >> the transport drivers need to be aware of their own maximum frame sizes >> and allocate accordingly. Ie totemrrp_malloc() doesn't need a frame >> size parameter, (but instead totemudp.c would allocate a frame size that >> is appropriate for the transport (in the case of udp, something like 10k >> would be safe). > > Cool, can do. Thanks for the input. > >> Another interesting idea that may be useful for a second patch is to >> keep a list of freed frames to avoid the malloc/free overhead for the >> common cases of totemsrp. Surprisingly this is quite a significant >> amount of overhead (according to oprofile). >> >> A totemnet_malloc operation would then be something like: >> if free_list has a frame, return frame and delete frame from free list >> for the instance >> else malloc frame using transport driver >> totemnet_free operation would be something like: >> add frame to free list >> >> To do this "slabbing" properly, a conditional parameter is probably >> needed in the transport driver definition because totemnet_free for the >> iba transport actually needs to call the free function (so it can be >> reposted to the send queue). > > Yes, or perhaps push it down into the drivers and have them call a common > implementation of the free list? I guess one downside of that is that you end > up with a separate pool for each driver. I'll keep it in mind; as you say, > it's probably a separate patch. > > cheers, > Zane. > >> Regards >> -steve > _______________________________________________ Openais mailing list [email protected] https://lists.linux-foundation.org/mailman/listinfo/openais
