100 likes | 129 Views
SIP Overload Control IETF Design Team Status. Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent. SIP Overload Control Design Team Current Status. Team Members New design team members Ahmed Abd el al, Tom Phelan (Sonus Networks) Existing team members
E N D
SIP Overload ControlIETF Design Team Status Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent
SIP Overload Control Design TeamCurrent Status • Team Members • New design team members • Ahmed Abd el al, Tom Phelan (Sonus Networks) • Existing team members • Eric Noel, Carolyn Johnson (AT&T Labs) • Volker Hilt, Indra Widjaja (Bell Labs/Alcatel-Lucent) • Charles Shen, Henning Schulzrinne (Columbia University) • Mary Barnes (Nortel) • Jonathan Rosenberg (Cisco) • Nick Stewart (British Telecom) • Four independent simulation tools • AT&T Labs, Bell Labs/Alcatel-Lucent, Columbia University, Sonus Networks • Draft submitted • draft-hilt-sipping-overload-design-00
draft-hilt-sipping-overload-design-00 Design considerations and models for an overload control solution for SIP. • Frame the discussion of an SIP overload control mechanism. • Describe possible design choices and models. • This is NOT a proposal for a solution! Contributors are the members of the SIP overload control design team.
draft-hilt-sipping-overload-design-00Implicit/Explicit Overload Control Implicit Overload Control • Goal is to achieve a gentle decline in throughput as overload increases without explicit overload feedback. • SIP server overload is regenerative. • Messages that get dropped due to overload are retransmitted and increase the offered load for the already overloaded server. • Selecting messages to be dropped during overload requires processing at a SIP server. • Resources spend on parsing messages to identify new requests that can be discarded during overload are lost. • The number of successful transactions declines as load increases, even with fully non-regenerative overload behavior. Explicit Overload Control • An explicit overload signal is used to request a reduction in the incoming load. • Enables a SIP server to adjust the incoming load to a level at which it can perform at maximum capacity.
C A A B D D B D … C Hop-by-hop Client-to-Server Server-to-Server b a z c x C A B D x End-to-end draft-hilt-sipping-overload-design-00Topologies/Degree of Cooperation • Client-to-Server vs. Server-to-Server • Hop-by-hop vs.End-to-end
draft-hilt-sipping-overload-design-00Overload Control Feedback Types Rate-based Overload Control • Idea: limit the request rate at which an upstream element is allowed to forward requests to the downstream neighbor. • A server instructs each upstream neighbor to send at most X requests per second. Loss-based Overload Control • Idea: reduce the number of requests an upstream element would normally forward to the downstream neighbor by a percentage X. • A server instructs each upstream neighbor to reduce load by X%. Window-based Overload Control • Idea: allow an entity to transmit a certain number of messages before it needs to receive a confirmation for the messages in transit. • An upstream neighbor maintains an overload window that limits the number of messages in transit without being confirmed.
draft-hilt-sipping-overload-design-00Next Steps Is this document useful to the SIPPING WG?
SIP Overload Control Design TeamSimulation Models (1) Different Network Conditions • Evaluate overload control mechanisms under varying network conditions. • Different network loss probabilities. • Different transmission delays. • Combination of transmission delay and loss probability. Offered Load Distribution • Evaluate the impact of an uneven distribution of load on a network of SIP servers. • Fraction 1-f of edge servers operate at engineered load, fraction f operate above engineered load. • Examine steady-state as well as transient behavior.
SIP Overload Control Design TeamSimulation Models (2) Overload Events • Evaluate the transient behavior of SIP servers when overload occurs/is removed. • Scenarios: • Sudden peak • Sudden peak with gradual release • Temporary, light overload Changes in Neighbor Set • Evaluate how well an algorithms adapts to senders starting/stopping to transmit.
SIP Overload Control Design TeamFairness A criteria to evaluate overload control algorithms is how well they fulfill a given fairness criteria: Basic user-centric fairness • Definition: equal success probability for each individual user • Example: “Third caller receives free tickets” Basic provider-centric fairness • Definition: equal success probability for each provider • Example: specific SLAs Customized user-centric and provider-centric fairness • Definition: any fairness requirements other than the basic ones. • Examples: • Specific source sending an unusually high load is throttled more than other sources. • A SLA requires specific (un-equal) sharing of the capacity among upstream servers.