Operator LB entropy hash changes without a stable address

[Gyan Mishra <[e-mail]>]

Fernando

Can we add txt related to this topic of L2 Ethernet bundle hash or L3 multi
parallel links ECMP hash changes when temporary address is employed.  This
is commonly used by all operators:

L2 hash load balancing hash entropy:
802.1ad is a QinQ  technology used by providers for backbone bridges
Ethernet relay services instances.  Inner tag is provider label and outer
tag is customer label.  The flow hash entropy is based on
source/destination hash, however if the address is not stable and changing,
that changes the hash entropy for load balancing flows.  That throws off
the lb entropy and now you don’t get close to 50/50 lb entropy with 2 links
in an Ethernet bundle. There are many instances when bundled ethernet
aggregation or MLAG is employed where a Src/dest hash is used to create
entropy for load balancing of flows ; as well as in MPLS L2 vpn  and NG L2
VPN ethernet service instance e-line, e-lan, e-tree where a fat pseudo
wire” is deployed requiring stable IP for lb entropy to work properly.

L3 ECMP hash load balancing entropy:
Another instance where stable IP is necessary is related along those same
lines of thought to BGP IGP ISIS or OSPF ECMP flow based load balancing
hash hash where flows lb entropy is created to get the even load balancing
over multiple parallel L3 links in a enterprise or provider operators
network.  When the address keeps changing that changes the load balancing
entropy for the hash.  So now instead of the flow going over L3 link A it
is now going over link B.

So this is an operational impact consequence on a operators network
enterprise or service provider from address changing.  Thinking about this
further if let’s say all addresses change based on current valid 9 days /
preferred 1 day you have a new preferred address daily.  So let’s say their
are thousands of flows on a particular L2 or L3 path the hash is constantly
changing.  It’s possible with the randomness off all the addresses
constantly changing some may or may not rehash to a different path so load
balancing may or may not be impacted.  However the larger impact is now
troubleshooting an operators network to trace a flow and determine which
link is carrying a particular flow which is now ever changing.  So added
MTTR (mean time to recovery) during an outage for operations to debug and
troubleshooting  tracing a flow.
-- 

Gyan  Mishra

Network Engineering & Technology

Verizon

Silver Spring, MD 20904

Phone: [phone]

Email: [e-mail]