Thomasian (1987) refined this approach. Queries are classified not only into CPU-
or I/O-bound, but into a finite set of query types based on all their service demands
(i.e., CPU, memory, different disks). Three routing algorithms have been compared
in a simulation study. The results again showed that query-dependent routing based
on query classification significantly improves the system performance. The article
also concludes that an accurate response time prediction algorithm is quite complex
and that the mapping of incoming queries into a fixed set of query types remains
difficult to find and to achieve.
Affinity-Based Transaction.Routing
The idea behind affinity-based routing is to assign queries that access the same data
to the same component. These are typically query-dependent strategies. For example,
Yu, Cornell, Dias, and Iyer (1987) presented an affinity-based approach to transaction
routing that classified incoming transactions into affinity groups based on previous
knowledge of the database call reference pattern. The reference pattern is retrieved
from trace information of the transaction workload. The proposed routing scheme
assigns a fixed destination system to each affinity group. In simulation study, the
authors showed that affinity-based routing significantly reduced lock-contention and
buffer I/O, and therefore clearly improved the response times of transactions.
Pages:
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454