Distributed consensus

PDF Publication Title:

Distributed consensus ( distributed-consensus )

Previous Page View | Next Page View | Return to Search List

Text from PDF Page: 060

60 3.5. PHASE ORDERING §4.2][OO14, §5.1][VRA15, §3][MPSP10, §3], usually in combination with Multi-Paxos (dis- cussed in §3.6). 3.5 Phase ordering During phase one of Classic Paxos, the proposer does not require knowledge of the value γ which they will propose if possible, in phase two. It is therefore possible for a proposer to execute phase one prior to knowledge of a value to propose. When the proposer then learns the value to propose, it may now decide this value in one round trip instead of two, provided no other proposer has also executed the proposer algorithm for a greater epoch. We can increase the likelihood that this will occur by also making this proposer a distinguished proposer. This observation is widely utilised [Lam01a, §3][MPSP10, §4], usually in combination with a distinguished proposer and Multi-Paxos (discussed in §3.6) 3.6 Multi-Paxos Thus far, we have considered how to reach consensus over a single value. In practice, these algorithms are usually utilised to reach consensus over an infinite sequence of values. Broadly speaking, we can divide existing algorithms for consensus over a sequence into two families: Classic Paxos algorithms which are based upon executing multiple distinct instances of single valued consensus. Examples include Classic Paxos, Mencius [MJM08] and Fast Paxos [Lam05a]. These approaches are rarely used in production systems. Multi-Paxos algorithms where one proposer takes the role of leader by executing phase one over the sequence and then coordinates decisions until a new leader takes over. This approach is widely utilised in production systems. Examples include Chubby [Bur06, CGR07], Zookeeper [HKJR10, JRS11], Ring Paxos [MPSP10], View- stamped Replication [OL88, LC12] and Raft [OO14]. Multi-Paxos is an optimisation of Classic Paxos for consensus over a sequence. Multi-Paxos differs from successive instances of Classic Paxos in one key way. The phase one of Classic Paxos is shared by all instances. Each acceptor needs only to store the last promised epoch once. Prepare and promise messages are not instance-specific and therefore do not need an index included in the phase one messages. This is combined with distinguished proposer (§3.4) and phase ordering (§3.5) optimisations as follows. Phase one is executed by a proposer prior to knowledge of the values to propose.

PDF Image | Distributed consensus

PDF Search Title:

Distributed consensus

Original File Name Searched:

UCAM-CL-TR-935.pdf

DIY PDF Search: Google It | Yahoo | Bing

Cruise Ship Reviews | Luxury Resort | Jet | Yacht | and Travel Tech More Info

Cruising Review Topics and Articles More Info

Software based on Filemaker for the travel industry More Info

The Burgenstock Resort: Reviews on CruisingReview website... More Info

Resort Reviews: World Class resorts... More Info

The Riffelalp Resort: Reviews on CruisingReview website... More Info

CONTACT TEL: 608-238-6001 Email: greg@cruisingreview.com (Standard Web Page)