logo

Distributed consensus

PDF Publication Title:

Distributed consensus ( distributed-consensus )

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

Text from PDF Page: 145

CHAPTER 8. CONCLUSION 145 In section 7.4, we proposed a hybrid algorithm consisting of using an allocator for the minimum epoch and pre-allocation for all other epochs. Provided the allocator is live and synchronous, then any number of proposers will terminate in two rounds trips (one to the allocator, one to the acceptors). 8.3.3 Improved performance Our generalisations provide the opportunity to improve the best case performance and/or to increase the likelihood of the best case occurring in practice. Optimising for the steady state has allowed us to improve overall performance. The tradeoff for this gain can be decreased performance during the rarer failure case behaviour. Unlike Classic Paxos, we do not enforce a particular tradeoff between performance and availability. Instead, this tradeoff is an application specific decision. This is best illustrated by combining Multi-Paxos with the weakened quorum intersection between phases (§4.1). We can now choose our tradeoff between phase one quorums, which are rarely used as they are only needed when replacing a leader and phase two quorums, which are used for every decision. The key motivation for Multi-Paxos is to reach agreement in one round trip, however, its centralised approach is a substantial performance bottleneck. We have proposed various other mechanisms to achieve one round trip agreement without centralisation, including the following: If proposers and acceptors are co-located on each participant: • A participant can execute phase one locally, provided it uses all participants for phase two (§4.1). • A participant can complete phase one locally, provided the participant has accepted a proposal from the predecessor epoch (§5). Otherwise: • One of the proposers can skip phase one if it has been allocated of the minimum epoch (§4.2.3). When reaching agreement over a sequence, this proposer can be rotated to avoid centralisation. • A proposer can skip phase one to propose its candidate value if its candidate value is assigned to the minimum epoch using epoch allocation by values (§7.2) • Any proposer can skip phase one to propose its candidate value if the minimum epoch is assigned by epoch allocation by recovery (§7.3)

PDF Image | Distributed consensus

distributed-consensus-145

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 | RSS | AMP