Update keps/sig-network/4963-kube-proxy-services-acceleration/README.md

Co-authored-by: Adrian Moisey <adrian@changeover.za.net>

Author: 2 people

keps/sig-network/4963-kube-proxy-services-acceleration/README.md

@@ -176,7 +176,7 @@ A configurable parameter (--offload-packet-threshold) will determine the minimum
 
 The default value for the --offload-packet-threshold will be carefully chosen based to ensure optimal performance for a wide range of applications. The traffic that will get more benefit from this feature will be the so called [elephant flows](https://en.wikipedia.org/wiki/Elephant_flow), so we'll obtain our default value based on that.
 
-The [elephant flow detection is a complex topic with a considerable number of literature about it](https://scholar.google.pt/scholar?q=elephant+flow+detection&hl=es&as_sdt=0&as_vis=1&oi=scholart). For our use case we proposa a more simplistic approach based on the number of packets, so it can give us good trade off between performance improvement and safety, we want to avoid complex heuristics and have a more predictible and easy to think about behavior based on the lessones learned from [KEP-2433 Topology aware hints](https://github.com/kubernetes/enhancements/tree/master/keps/sig-network/2433-topology-aware-hints#proportional-cpu-heuristic).
+The [elephant flow detection is a complex topic with a considerable number of literature about it](https://scholar.google.pt/scholar?q=elephant+flow+detection&hl=es&as_sdt=0&as_vis=1&oi=scholart). For our use case we propose a more simplistic approach based on the number of packets, so it can give us good trade off between performance improvement and safety, we want to avoid complex heuristics and have a more predictable and easy to think about behaviour based on the lessons learned from [KEP-2433 Topology aware hints](https://github.com/kubernetes/enhancements/tree/master/keps/sig-network/2433-topology-aware-hints#proportional-cpu-heuristic).
 
 We chose 20 as the number of packets as the threshold to offload, based on existing thresold used by Cisco systems in [Cisco Application Centric Infrastructure](https://scholar.google.com/scholar_lookup?hl=en&publication_year=2018&author=G.+Tam&title=Cisco+Application+Centric+Infrastructure), Cisco referred to an elephant flow if the flow contains more than 15 packet sizes, i.e., short flow is less than 15 packets, we add 5 packets of buffer to be on the safe side. This means that, using TCP as an example, and assuming an MTU of 1500 bytes and removing the overhead of the TCP headers (that can vary from 20-60 bytes, use 40 for this example), offloading will benefit workloads that transmit more than: TCP Payload * Number of packets = 1460 bytes/packet * 20 = 29200 bytes.