Other Coordination Patterns
Suppose processes are cooperating in a search to find a value, xmax, that maximizes a function F; that is, F(xmax) is the greatest value F attains. Each process has its own list of x values to test. We would like to do something like:
for x in MyCandidateList: currentMax = ws.fetch('max') y = f(x) if y > currentMax: ws.store('max', y)
But that would be wrong, since fetch consumes a value that may not be replaced.
We would like a way to consult a variable without destroying the value. In NWS, find returns a value without destroying it. So, we could try:
for x in MyCandidateList: currentMax = ws.find('max') y = f(x) if y > currentMax: ws.store('max', y)
But that would be wrong toowe are no longer maintaining a single maximum value, and currentMax might not really be current. How about:
for x in MyCandidateList: currentMax = ws.find('max') y = f(x) if y > currentMax: currentMax = ws.fetch('max') if y > currentMax: currentMax = y ws.store('max', currentMax)
Now we're only updating max when our currentMax really is larger. Because of the fetch/store pair, we have just one value associated with max (this pair of operations also ensures the atomicity of the update). But suppose it is unlikely that a given x will result in a new maximum. We could further reduce the coordination traffic by invoking find only every other iteration or every tenth or... Because we always consult the true current value of max before committing to a change, stale values won't lead to incorrect results. They will, however, increase the number of failed update attempts.
There are other uses for find, the most common being "write-once" variables such as initialization data that are set at the beginning of a computation and needed by two or more ensemble members.
find alters the way value queues are referenced. What about variations in the queue itself? NWS supports four modes for values:
- First-In-First-Out (the default).
- Nondeterministic (not random, just arbitrary).
- Single (the variable only holds one value; store overwrites the previous value, if any).
The single mode works well with the find operation and is useful for status values; you can update the status without bothering to remove (fetch) the previous status.
from nws.client import SINGLE ws.declare('status', SINGLE) while True: time.sleep(60) status=... ws.store('status', status)