Re: [ovs-dev] [PATCH] Simplify kernel sFlow implementation

2011-08-11 Thread Ben Pfaff
The 64-bit packed field is just an intermediate step. I agree that it is somewhat ugly. It will be replaced by a unique ID of one kind or another (an incrementing counter? random number?) either when it gets to be too small, or when we decide to fix up the stats race window I described earlier.

Re: [ovs-dev] [PATCH] Simplify kernel sFlow implementation

2011-08-11 Thread Neil McKee
OK. Makes sense. It might be OK from an sFlow perspective if the user-space lookup occasionally fails.Just as long as it never retrieves the wrong action-list. That would be a problem. If you can look up the actions in user-space, do you still need to generate that 64-bit packed-field?

Re: [ovs-dev] [PATCH] Simplify kernel sFlow implementation

2011-08-11 Thread Ben Pfaff
Neil, don't worry! This really is a step forward and not a step back. The actions passed to the kernel and back to userspace are not really as helpful as you think, because they discard a lot of semantic information. These actions do say *what* happened but they do not say *why*. One off-the-cuf

Re: [ovs-dev] [PATCH] Simplify kernel sFlow implementation

2011-08-11 Thread Neil McKee
I think the current behavior where the whole actions-list is serialized and passed up with the sample is actually really elegant and future-proof. New actions are being added all the time, and having access to a sampled-stream of "exactly-what-is-happening-in-there" is always going to be good

Re: [ovs-dev] [PATCH] Simplify kernel sFlow implementation

2011-08-09 Thread Ben Pfaff
On Wed, Aug 10, 2011 at 01:18:35PM +0800, Jesse Gross wrote: > The rough goal that I was thinking of (although certainly without all > of the details fleshed out) is to have a sample action which takes a > probability and sublist of actions which is executed according to the > probability. The exi

Re: [ovs-dev] [PATCH] Simplify kernel sFlow implementation

2011-08-09 Thread Jesse Gross
On Wed, Aug 10, 2011 at 12:42 PM, Ben Pfaff wrote: > On Tue, Aug 09, 2011 at 10:46:38AM -0700, pravin shelar wrote: >>  This patch is first step toward generic/simple sampling >>  action which will be completely independent of sflow. >>  Following patch cleanup sampling upcall by striping it down

Re: [ovs-dev] [PATCH] Simplify kernel sFlow implementation

2011-08-09 Thread Ben Pfaff
On Tue, Aug 09, 2011 at 10:46:38AM -0700, pravin shelar wrote: > This patch is first step toward generic/simple sampling > action which will be completely independent of sflow. > Following patch cleanup sampling upcall by striping it down to > minimum essential data that userspace can not figur

[ovs-dev] [PATCH] Simplify kernel sFlow implementation

2011-08-09 Thread pravin shelar
This patch is first step toward generic/simple sampling action which will be completely independent of sflow. Following patch cleanup sampling upcall by striping it down to minimum essential data that userspace can not figure out for itself. Signed-off by: Pravin Shelar --- datapath/action