Tools
Sorted by:
Try your query at:
 |
 |
 |
 |
 |
 |
Results 1 - 10
of
236,720
Table 1. Consensus algorithms in the crash-recovery model
2005
"... In PAGE 8: ... Fast Paxos writes once per round, as does the algorithm in [17]. Table1 summarizes consensus algorithms in terms of communication steps (i.e.... In PAGE 8: ... Notice that we consider messages sent from a process to itself, as these messages also impose some processing cost at each machine. From Table1 , the optimized version of Paxos takes the same number of com- munication steps as B*-Consensus but, due to its centralized nature, needs nearly half the messages. Two more communication steps are required when Paxos runs the first phase.... ..."
Cited by 2
Table 3: Data manager communication during crash recovery. CLASSICAL PRESUMED EARLY COORD.
1993
"... In PAGE 35: ... 8.2 Crash Recovery Considerations Table3 presents the number of coordinators a crashed DM must contact when it recovers. The rst row contains the number of coordinators a crashed DM must contact before it can begin executing any new transaction.... ..."
Cited by 16
Table 4: Coordinator communication during crash recovery. CLASSICAL PRESUMED EARLY COORD.
1993
"... In PAGE 36: ... These coordinators may be contacted at any time. Table4 considers the case of a crashed coordinator. M is the total number of DMs in the system.... ..."
Cited by 16
Table 2.1. Cushion-suction effect. An example of counter-measures to negative side-effects is maneuvering during a crash stop. A crash stop can be effected by setting the propeller in the reverse. This creates a turning moment which can be compensated by rudder.
Table 2 presents the failure-free overhead of different logging protocols. It lists total execution time, the mean log size, the total log size, and the number of times the volatile logs are flushed to stable storage. Since no logging exists in the home-based TreadMarks, its execution time serves as a performance baseline. Both our coherence-centric logging (CCL) and traditional message logging (ML) protocols pro- vide home-based SDSM with crash recovery capabilities, but they involve different performance penalty amounts.
"... In PAGE 6: ... Table2 . Overhead Details under Different Logging Protocols.... In PAGE 6: ... Overhead Details under Different Logging Protocols. From Table2 , it is apparent that our CCL consistently results in lower failure-free overhead than ML, which in- duces a larger mean log size and is affected significantly by high disk access latency. CCL keeps a far less amount of logged data because it stores only coherence-related infor- mation that cannot be retrieved or recreated after a failure, whereas ML simply records all incoming messages.... ..."
Table 3: The crash recovery time for difierent num- bers of log records lost in a system crash (i.e., BS)
2006
Cited by 1
Table 1. Replicated JMS server: comparison between replicated and non-replicated context. Non-replicated context Replicated context
"... In PAGE 10: ... 4.3 Comparison of the JMS server replication types Table1 presents the comparison between the non-replicated context and replicated context solutions. The replicated context solution uses the simpler crash-stop failure model, has an option not to use the stable storage and most importantly does not isolate the clients in the case of a server replica crash, which greatly improves system liveness.... ..."
Table 2.2. Crash stop maneuver. If our hypothesis is correct, then the system change should lend itself to a similar analysis. Let us start by filling out the problem part of the form. See Table 2.3.
Table 2.2. Crash stop maneuver. If our hypothesis is correct, then the system change should lend itself to a similar analysis. Let us start by filling out the problem part of the form. See Table 2.3.
Results 1 - 10
of
236,720
