[j-nsp] QFX5100 issues with IPv6 hashing

[Anton Yurchenko]

Hi,

I am using the -P option for iperf, it does vary source port dst port is 
the same though. Normally (and with v4) traffic it results in good 
hashing behavior.

Thanks,

On 4/18/15 07:21, Michael Loftis wrote:
>
>
> On Friday, April 17, 2015, Anton Yurchenko <ayurchenko at gmail.com 
> <mailto:ayurchenko at gmail.com>> wrote:
>
>     Hi All,
>
>     I am seeing issues with QFX5100, hashing of IPv6 traffic over ECMP
>     paths is very bad.
>     In my case I have an ECMP over 3x40Gig links, and sending
>     180x10Mbit UDPv6 flows over them. First 40G gets ~300mbit, second
>     ~600 and third ~900.
>     IPv4 hashing is fine ~5% variance in traffic levels.
>
>     I have enabled hashing on L3/L4 and layer2-payload but no avail.
>
>     It seems that protocol/ports are not taken into account at all. If
>     I start traffic from just one IP with 30 flows, they will all end
>     up on a single link.
>
>
> You say 30 flows but are they differing at all? Even for v6 it looks 
> at Source IP, Dest IP, Proto, Source port, Dest port. If you traffic 
> generator doesn't vary the source port then each "flow" will look 
> identical to the hash algo.
>
>
>     Seeing this on 13.2X51-D35.3 and 14.1X53-D25.2 releases.
>
>     Here is my hashing config:
>
>     load-balance {
>         indexed-load-balance;
>     }
>     hash-key {
>         family inet {
>             layer-3;
>             layer-4;
>         }
>     }
>     enhanced-hash-key {
>         hash-mode {
>             layer2-payload;
>         }
>     }
>
>     Wondering if anybody have seen something similar or have seen it
>     working properly.
>
>     Thanks!
>     _______________________________________________
>     juniper-nsp mailing list juniper-nsp at puck.nether.net
>     https://puck.nether.net/mailman/listinfo/juniper-nsp
>
>
>
> -- 
>
> "Genius might be described as a supreme capacity for getting its 
> possessors
> into trouble of all kinds."
> -- Samuel Butler
>