richlowe: I thought it was "ask one of them what the other one would say, then do the opposite".

but the real answer is "define a mechanism to extract a per-connection flow id from each packet; and hash it to pick one of the paths"

the fun part though is when you have multiple routes for your DNS servers, but only some of them actually work

hello

sommerfeld: the counter to hashing something about the packet to choose the route is sys admins who don't understand that's what LACP does and you have to configure it based on the workload - I've had a few instances of sysadmins complaining their single stream iperf test between two systems with LACP trunk tops out at just under the speed of a single link.

usual result is LACP is hashing on something like mac-address

that's different from LACP

and yes, that's pretty normal with aggrs

because if you take a tcp stream and start spreading it across multiple links, there's a good chance they won't arrive in order, and you're likely to lose more performance in the reassembly overhead...

i think i've asked this before, but i still have this enhancement to prtconf: gist.github.com/jasonbking/52f11d7104af80546f222dd7296249c9 -- would there be interest in throwing it up for review?

jbk: It's not clear what one should be looking at in that gist.

jbk: what's the thing you're changing? (at a first glance it looks like prtconf -v output)

the decoding of the address ranges

as well as including b/d/f values

e.g. lines 210-215 or 874-876

(it's just decoding the property values and not issuing any extra ioctls or such)

since the current address output for pci devices is wrong (and it seems whenever PCI-LPC bridges became a thing, it was never updated to decode the ISA data behind an LPC bridge (I'm guessing in the 486/pentium days, the isa nexus hung off the root nexus?)

(lines 2924-2949)

(it annoyed me, so I wanted to do something about it :P even if just for me)

I'm always up for more decoded stuff!

jbk: the "isa nexus" works two ways _now_

jbk: search for references to `pseudo_isa`

yeah, i'm aware of that.. i adapted some of it for hyper-v where there's no LPC bridge

and you just need to instantiate one for the ACPI stuff (e.g. serial ports)