There are two principally different options to make this fence:
(A) by distributing tasks;
(B) by making all tasks on site sequentially.
In case (A), distributing task scheme assumes the existence of agents–workers and
distributers and their abilities to act:
– N workers for plank processing are available and ready;
– a distributor of the nails is in place;
– a distributer of the hammers is in place;
– a distributer of the planks is in place;
– a distributer of rails is in place;
– a collector of the fence segments initially is and placing the planks;
– nailing the planks at two rows are performed by workers;
– collecting the hammers is performed; and
– garbage collector is in place and completes the task execution.
Case (B), in turn, assumes that the same worker is doing all actions, like “a jack for
all trade”, has one hammer, bucket of nails, and does the following:
– takes nails;
– planks where they are;
– half-nail planks;
– places them on the rails;
– nails them all;
– place fence where necessary, collect garbage.
Let us consider the process of making the fence from N planks in more details for
both cases, assuming that nails, hammers, planks, and rails are ready and placed in
the local warehouse (storage and executed by “a system officer”, while workers
execute user task). Sequences are presented in Table 17.7.
Our task now is about giving elementary time slot t e and constant coefficients
equal for both variants of fence processing to prepare two variants of the fence
completion as a sequence of steps for A and B cases. This will illustrate a gain from
distribution of works.
We need to compare these cases as well as explain what is possible to prepare in
preprocessing and what is possible only during operation. One might find useful to
make a table of all works mentioned and using own experience and case estimate a
concrete gain for concrete case.
Now we have to answer the following questions:
When distributed computing is efficient in comparison with sequential;
What impact system software makes on parallelization of task and efficiency of a
system.
It is clear that planks are data, nails and hammers are programs to process data
on site, and distributer is runtime system.
244
17 On Performance: From Hardware up to Distributed Systems
(A) by distributing tasks;
(B) by making all tasks on site sequentially.
In case (A), distributing task scheme assumes the existence of agents–workers and
distributers and their abilities to act:
– N workers for plank processing are available and ready;
– a distributor of the nails is in place;
– a distributer of the hammers is in place;
– a distributer of the planks is in place;
– a distributer of rails is in place;
– a collector of the fence segments initially is and placing the planks;
– nailing the planks at two rows are performed by workers;
– collecting the hammers is performed; and
– garbage collector is in place and completes the task execution.
Case (B), in turn, assumes that the same worker is doing all actions, like “a jack for
all trade”, has one hammer, bucket of nails, and does the following:
– takes nails;
– planks where they are;
– half-nail planks;
– places them on the rails;
– nails them all;
– place fence where necessary, collect garbage.
Let us consider the process of making the fence from N planks in more details for
both cases, assuming that nails, hammers, planks, and rails are ready and placed in
the local warehouse (storage and executed by “a system officer”, while workers
execute user task). Sequences are presented in Table 17.7.
Our task now is about giving elementary time slot t e and constant coefficients
equal for both variants of fence processing to prepare two variants of the fence
completion as a sequence of steps for A and B cases. This will illustrate a gain from
distribution of works.
We need to compare these cases as well as explain what is possible to prepare in
preprocessing and what is possible only during operation. One might find useful to
make a table of all works mentioned and using own experience and case estimate a
concrete gain for concrete case.
Now we have to answer the following questions:
When distributed computing is efficient in comparison with sequential;
What impact system software makes on parallelization of task and efficiency of a
system.
It is clear that planks are data, nails and hammers are programs to process data
on site, and distributer is runtime system.
244
17 On Performance: From Hardware up to Distributed Systems