With those approaches, a socalled
freshness index f(d) ??? [0, ..., 1] measures the freshness of some data d. This
freshness index reflects how much the data has deviated from the up-to-date version.
Intuitively, a freshness index of 1 means the data is up-to-date, while an index
of 0 would characterise the data as ???infinitely??? outdated. One can distinguish three
basic approaches to define the freshness index: delay freshness, version freshness,
and data deviation.
??? Delay.freshness: A delay freshness index reflects how late a certain cluster
node is as compared to the up-to-date OLTP node (Guo et al., 2004; R?¶hm et
al., 2002). It is based on the period of time between the last propagated update
and the most recent update on the up-to-date node. Let ?„ (c) denote the
commit time of the last refresh subtransaction on an OLAP node c, and ?„ (c0)
the commit time of the most recent update subtransaction on the OLTP node.
Then the delay freshness index is defined as f (c) = ?„ (c) / ?„ (c0). This implies
that f ??? [0, 1].
??? Version.freshness: Alternatively, one can base the freshness definition on the
version difference between a cluster node and the up-to-date node c0. We define
the version of a cluster node c as the number of committed update transactions,
denoted by v(c). Following Pacitti and Simon (2000), the version freshness
index is then defined as f(c) = v(c) / v(c0) ; f ??? [0, 1].
Pages:
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462