Myra Spiliopoulou and Johann Christoph Freytag. Modelling the dynamic evolution of system workload during pipelined query execution. Technical Report ISS-20, Institut fur Wirtschaftsinformatik, Humboldt-Universitat zu Berlin, Germany, 1995. http://www.wiwi.hu-berlin.de/institute/iwi/info/research/iss//papers/ISS20.ps.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of different operators from the 5 same execution schedule is modelled, taking into account pipelined and partitioned execution. The approach is applied to a particular system (DB2 Parallel Edition on an IBM SP2 architecture) but no comparison results are reported. M. Spiliopoulou et al. [30,28,29] study the problem of estimating execution time for queries composed of multiple pipelined operators scheduled to run in a parallel database system. The intention is to incorporate the execution time prediction mechanism into a generic optimiser for parallel query processing. The developed cost ....

M. Spiliopoulou and J. C. Freytag, "Modelling the dynamic evolution of system workload during pipelined query execution", Technical Report ISS-20, Institut für Wirtschaftsinformatik, Humboldt-Universität zu Berlin, Germany, 1995.

....A pipeline executes at the pace of its slowest process. Due to latency, a process affects the execution pace of the pipeline only when it becomes runnable. Therefore, the execution rates of pipelined processes must be synchronized on the fly. We present such a dynamic synchronization mechanism in [5]: We compute the relative execution rates of all producers of a node x, and identify the slowest producer y 1 . We specify two adjustment functions taking values in the (0; 1] range: adjustConsumer(x; y 1 ) slows down the consumer, while adjustP roducer(x; y i ) reduces the execution speed of ....

....of the time span C (x; S x ) Let y be its slowest producer of x. Then: x; S x ) x; S x ) x is a leaf (y; S y ) adjustConsumer(x; y) Delta C (x; S x ) otherwise (2) If x is too fast for its producers, it is slowed down by the adjustment function adjustConsumer( Delta) [5]. Let C (x; S x ) be the completion time of x, i.e. its remaining execution time after C (x; S x ) If the consumer of x runs on the same processor as x, then the contents of S x change into S x and the processor s resources must be redistributed. According to Eq.1: C(x; S x ) ....

[Article contains additional citation context not shown here]

Myra Spiliopoulou and Johann Christoph Freytag. Modelling the dynamic evolution of system workload during pipelined query execution. Technical Report ISS-20, Institut fur Wirtschaftsinformatik, Humboldt-Universitat zu Berlin, Germany, 1995. http://www.wiwi.hu-berlin.de/institute/iwi/info/research/iss//papers/ISS20.ps.

....However, due to latency, a process affects the execution pace of the pipeline only when it becomes runnable. Therefore, an explicit synchronization mechanism is needed, which adjusts the execution rates of pipelined processes on the fly. We present such a dynamic synchronization mechanism in [SF95]. In summary, this mechanism computes the relative execution rates of all producers y 1 , y k of a node x, and identifies the slowest producer, say y 1 . Depending on whether the consumer or the producers must be slowed down, two adjustment functions are specified: adjustConsumer(x; y 1 ) ....

....(x; n 0 x ) We compute it as follows: T end (x; n 0 x ) adjustProducer(z; x) Delta C end (x; n 0 x ) T init (x; n x ) 3) where the adjustment function adjustProducer( Delta) slows down x if it is faster than z; otherwise it is equal to 1. This function is described in detail in [SF95]. 3.3 Example In Fig. 5, we consider a simple QEP scheduled on one processor. The cost of this schedule is the execution time of x and the elapsed time until it starts execution. This elapsed time is equal to the initialization time of y 1 , assuming that y 1 is slower than y 2 . Note that C ....

[Article contains additional citation context not shown here]

Myra Spiliopoulou and Johann Christoph Freytag. Modelling the dynamic evolution of system workload during pipelined query execution. Technical Report ISS-20, Institute of Information Systems, Humboldt University of Berlin, Berlin, Germany, 1995. In preparation.

....A pipeline executes at the pace of its slowest process. Due to latency, a process affects the execution pace of the pipeline only when it becomes runnable. Therefore, the execution rates of pipelined processes must be synchronized on the fly. We present such a dynamic synchronization mechanism in [5]: We compute the relative execution rates of all producers of a node x, and identify the slowest producer y 1 . We specify two adjustment functions taking values in the (0; 1] range: adjustConsumer(x; y 1 ) slows down the consumer, while adjustP roducer(x; y i ) reduces the execution speed of ....

....S x ) Let y be its slowest producer of x. Then: T init (x; S x ) ae C init (x; S x ) x is a leaf T init (y; S y ) adjustConsumer(x; y) Delta C init (x; S x ) otherwise (2) If x is too fast for its producers, it is slowed down by the adjustment function adjustConsumer( Delta) [5]. Let C end (x; S x ) be the completion time of x, i.e. its remaining execution time after C init (x; S x ) If the consumer of x runs on the same processor as x, then the contents of S x change into S 0 x and the processor s resources must be redistributed. According to Eq.1: C end ....

[Article contains additional citation context not shown here]

Myra Spiliopoulou and Johann Christoph Freytag. Modelling the dynamic evolution of system workload during pipelined query execution. Technical Report ISS-20, Institut fur Wirtschaftsinformatik, Humboldt-Universitat zu Berlin, Germany, 1995. http://www.wiwi.hu-berlin.de/institute/iwi/info/research/iss//papers/ISS20.ps.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC