On 09/28/2015 01:10 PM, Jason Gunthorpe wrote: > On Mon, Sep 28, 2015 at 11:36:11AM -0400, Doug Ledford wrote: > >>> Also broadcast could cause a unecessary reception event on the NICs of >>> machines that have no interest in this traffic. >> >> This is true. However, I'm trying to balance between several competing >> issues. You also stated the revamped multicast code was adding latency >> and dropped packets into the problem space. Sending over the broadcast >> would help with latency. However, I have an alternative idea for that... > > I think your original idea of broadcast immediately and deferred > optimal mlid lookup is the best *functionally* for every case - only > when you enter the very edge world of caring about timing does it make > any difference. > > Christoph's needs would probably be better served by giving some API > to control the mlid cache (ie the neightbour table is already 99% of > the way there). This would let some userspace component pre-load and > fix all relevant data and undesired cache activity simply can't add > jitter. So, I've taken Christoph's patch, added two of my own (just changed the comment and the #if statement so that we create groups on send-only joins, and upped the max send-only backlog queue). We'll leave it at that for 4.3 and try to address it more fully in 4.4. -- Doug Ledford <dledford@xxxxxxxxxx> GPG KeyID: 0E572FDD
Attachment:
signature.asc
Description: OpenPGP digital signature