Browse Prior Art Database

A method to distribute packets to different queue base on tenants on Openstack environment. Disclosure Number: IPCOM000248071D
Publication Date: 2016-Oct-24
Document File: 4 page(s) / 53K

Publishing Venue

The Prior Art Database


In the openstack environment, the network part will provide the vlan, GRE(Generic Routing Encapsulation) and vxlan. If we select the vxlan as the network part. It will create the vxlan tunnel between the openstack components. The physical NIC(Network Interface Card) address will be used as the src ip and dst ip in the vxlan tunnel. So all the traffic in all the tenants, the src ip and dst ip will be the same if they send to the same destination and from the same server. This will have a issue during the receiver NIC do the hash function. The NIC hash function based on the src ip, dst ip,src port and dst port. So if the src ip, dst ip and dst port is the same, the hash result will be not good even the src port will change in some situation. At last, the most traffic will send to the same queue.

Our idea is automatic assign the different ip to vxlan tunnel as the src ip and dst ip in different tenants. There will be a ip address pool, each new flow will request two ip address from this pool. And the two new address will instead of the original ip in vxlan flow. So the src ip and dst ip in each new flow will different and it will help the receiver to hash the traffic to different queue.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 01 of 4

A method to distribute packets to different queue base on tenants on Openstack environment.

Openstack network is a typical SDN(Software Defined Network) network. In Openstack environment, all of packet usually was encapsulated by GRE or Vxlan protocol for support customer defined network.

If a packet was encapsulated by vxlan, all of network information, like source and destination ip address and src-tcp port and des-port will be hidden by tunnel.

So when a backend server received a packet, it will not award these network inform of packets before un-encapsulated packet. Sothe backend server will can't assign packet to different queue by base network information of 5 element (source ip, destination ip, source and destination port number and protocol number)

So all of packet will have to assign to a single queue.

RR can resolve this issue, but RR will introduce additional issue, tcp out of order.

Figure 1

Here is detail explain for packet was processed on Openstack environment On step 1, a packet was sent to public ip of network node from remote client


Page 02 of 4

On step 2, when network received this packet, it will check packet Infor and was processed by Vrouter and then this packet was forward to tunnel, the tunnel will add a vxlan header to this packet, finally the backend server received this packet.

The issue the switch and backend server will not recognized inner Infor of packet, because all of network Infor was hide by vxlan header. So the middle switch and backend will no method to split packet to different queue by l3 information
By assign different tunnel for each tenant, this disclosure can resolve above issue, all of packet belong to a certain tenant will assign to...