PDF Publication Title:
Text from PDF Page: 086
86 4.3. IMPLICATIONS 4.3.1 Bypassing phase two In section 3.2, we discuss how a Classic Paxos proposer can bypass phase two when a majority of acceptors return the same proposal (e, v) with promises in phase one. This is safe as (e, v) has already has been decided. The analogous optimisation is to return the value v when a phase two quorum Qe2 of acceptors return the proposal (e,v). This can result in not only skipping phase two but also skipping the remainder of phase one, if a Qe2 of acceptors return the same proposal before a Qf1 of acceptors return promises. 4.3.2 Co-location of proposers and acceptors In section 3.7, we discuss the option of co-locating both a proposer and an acceptor in each participant. We will now look at three algorithms which arise from combining this co-location with our weakened quorum intersection requirements. Example: All aboard Paxos One interesting implication of revision A that if we are willing to require all participants be up for guaranteed progress (and co-locate proposers and acceptors) then we can reach consensus in only one round trip. This is achieved by requiring all acceptors to accept in phase two. It is then sufficient under revision A for any acceptor to promise in phase one, as the intersection between the phases is still guaranteed. By co-locating the acceptors and proposers phase one can be completed locally without any communication with other participants. For example, in a system of 3 acceptors A = {a1, a2, a3}, the following are valid quorum sets: Q1 ={{a1},{a2},{a3}} Q2 ={{a1,a2,a3}} In contrast, under Classic Paxos we would still require intersecting quorums, such as majorities, for phase one so there is no advantage (only disadvantage) to requiring all acceptors to participate in phase two. Thus far, we have utilised revision A to achieve one round trip consensus provided all acceptors participated in phase two. The primary limitation of All aboard Paxos compared to Classic Paxos is that all participants must be live to guarantee progress instead of just a majority. We will now utilise revision B to overcome this limitation as follows. We will require all acceptors to accept in phase two for the epochs 0 to some epoch k. We will only require majorities to accept in phase two for all epochs from k + 1. Any value greaterPDF Image | Distributed consensus
PDF Search Title:
Distributed consensusOriginal File Name Searched:
UCAM-CL-TR-935.pdfDIY 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 |