1 / 6

IPv6 Flow Label for Server Load Balancing - update

IPv6 Flow Label for Server Load Balancing - update. draft-carpenter-flow-label-balancing-01 Brian Carpenter Sheng Jiang (Speaker) Willy Tarreau July/August 2012. Draft structure. The previous draft ( draft-carpenter-v6ops-label-balance) was split into two parts:

karyn-sullivan
Download Presentation

IPv6 Flow Label for Server Load Balancing - update

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. IPv6 Flow Label for Server Load Balancing - update • draft-carpenter-flow-label-balancing-01 • Brian Carpenter • Sheng Jiang (Speaker) • Willy Tarreau • July/August 2012

  2. Draft structure The previous draft (draft-carpenter-v6ops-label-balance) was split into two parts: 1. draft-carpenter-flow-label-balancing This describes use of the flow label for server load balancing, completely based on the existing standard (RFC6437). 2. draft-tarreau-extend-flow-label-balancing This proposes an extended model to support session persistence, requiring application layer changes. Not discussed here.

  3. Scenario Diagram IPv6 Clients in the Internet Ingress router Ingress router DNS-based←load splitting→ Possible flow label use Possible flow label use L3/L4 balancer L3/L4 balancer HTTP proxy … HTTP proxy TLS proxy TLS proxy … HTTPserver HTTPserver HTTPserver HTTPserver HTTPserver …

  4. Core of draft-carpenter-flow-label-balancing • According to RFC 6437, the flow label SHOULD be set to a suitable (uniformly distributed) value at source. • Behaviors on a stateful L3/L4 load balancer: • if flow label is not present, fall back to current methods • if flow label is present, but not a known value, apply the usual load balancing algorithm to select the server and remember the flow label <-> server mapping • if flow label is present and known, use it to select the proper server. Thus, for subsequent packets of flows, the load balancing can depend on 2 tuple {source address, flow label} – more efficient for ASICs than 5 tuple today • A stateless load balancer can also benefit by using 2 tuple as input of hash algorithm rather than 5 tuple

  5. Possible Benefits • Significant simplification for packets with IPv6 extension headers. • Assuming that 80-90% of IPv6 users reach the net without a proxy, large sites will be able to off-load most of their load balancing into ASIC-based LBs or even switches. • Ingress router sets flow label if zero • The remaining 10-20% of sessions will have persistence issues (multiple ports or source addresses) and will follow the normal route via the L7 LBs.

  6. Next steps fordraft-carpenter-flow-label-balancing The authors believe it describes a scenario that is consistent with existing standards and could be attractive in some deployments. Is this ready for Informational RFC?

More Related