Hence
we will assume delay freshness in the following discussion.
Overview.of.FAS
The basic mechanisms for ensuring serialisability and read consistency are as follows:
FAS uses the primary-copy replication scheme with deferred refreshment as
discussed before and executes updates first on the OLTP node. After an update subtransaction
finishes, as soon as a refresh is activated, FAS propagates the changes
to the remaining replicas using decoupled refresh subtransactions.
In more detail, some variant of multiversion concurrency control (MVCC) is used.
Freshness-aware scheduling does not provide just any version to a client but a version
which meets the given freshness limit. In other words, only cluster nodes with
freshness above the given lower bound will be considered during query routing.
Consequently, the higher the specified minimum freshness is, the smaller is the portion
of the cluster to which the corresponding query may be routed. In the worst case,
no node is available with the requested degree of freshness, and the coordination
middleware must activate update propagation first. Hence, although FAS follows
a lazy primary-copy replication approach with deferred updates, it nevertheless allows
queries to access the most recent data.
Example.3: Freshness-Aware Scheduling. Figure 3 shows three queries
with different freshness limits (R?¶hm et al.
Pages:
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464