river-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Patricia Shanahan <p...@acm.org>
Subject Re: datastructure classes
Date Tue, 14 Dec 2010 15:25:43 GMT
Carfield Yim wrote:
> Hi, copyonwritearraylist is suitable for this case imho. But it don't sound
> like an option to you. Would you share a bit more. I am interest to learn
> about it.

My first feeling was that there is at least one workload that seems
plausible to me for which it would have poor performance.

The workload is one in which most transactions are either a write or a
read-and-take. For example, consider a space that is the task pool for a
distributed workload. The writers are discovering tasks that need doing,
and recording them in the space. Each reader looks for a task it can do,
takes it from the space, and performs it. Now suppose the space is
buffering between bursty task creation and a fixed service rate.

At times, the queue will get long, because of a burst of new tasks being
added. Once that happens, copy-on-write becomes expensive.

Could people with more JavaSpaces experience comment on the plausibility
of this pattern?

I believe JavaSpace needs to be treated as an infrastructure that needs
robust performance across a wide range of workloads, not just
optimization to make one workload run fast. However, I am tempted to
implement a CopyOnWriteArrayList version, and say that any other
implementation must beat it on at least one benchmark.

Patricia

Mime
View raw message