03:41:46 <danmcd> @nomad    dladm show-aggr -p
03:41:52 <danmcd> Add that to gist
03:46:03 <danmcd> Also, did you try on a single ixgbe instead of aggrs?  (I wonder if the aggr code trying to find hash values might slow down on v6...)
03:46:55 <danmcd> If I had a cat6 cable I could replicate your experiment but I don't at the moment.
03:47:45 <danmcd> (Also it's release week for SmartOS.)
04:38:20 <nomad> danmcd, I tried it on a single previously, before I figured out IPv6 was the probable cause. I'll be tearing the aggregate apart next week and trying again.
04:39:12 <nomad> which fields do you want with that show-aggr? (It insists I have to give -o and arguments.)
04:41:01 <nomad> or did you want -P or --persistent ?
04:42:21 <nomad> https://pastebin.com/74wuHtnj (for dladm show-aggr --persistent)
15:12:13 <danmcd> I wanted -P nomad ==> balancing policy.
15:12:31 <danmcd> But if you're seeing similar degradation on a single link, it's better to run further checks there.
15:13:08 <danmcd> Also, what's the MTU for the link?  If it's not 9k (or whatever your ixgbe's max is) already, make it that way and see if the disparity matches or shrinks.
16:51:25 <sommerfeld> note that there will be a slight degradation in v6 line rate just due to the larger headers.   (20 bytes for ipv4 vs 40 for ipv6)
16:58:15 <danmcd> The differenecs he's quoting are way more than the 2% shrinkage on 1500MTU .