vasu@client3.cs.psu.edu (Vasu Nori) (03/02/91)
I am trying to define what a stable storage scheme for tuple space should be, in the face of process failures. Would anybody care to comment on the following : As far as concurrency control problem in Linda is concerned, it seems to be almost same as a distributed database concurrency control problem. Since the linda has fewer operations defined on the tuple space than does a database environment, achieving concurrency control seems simpler too. In fact, by regarding each tuple as an independent piece of information, (i.e., it doesn't directly affect any other tuple by its deletion or insertion into the tuple space) there is no need to define any serializability criteria across workers. Of course, this limits usage of linda to some applications. But probably it is not good for applications that require inter worker constraints (might introduce too many deadlocks anyway) So, in other words, linda is good for one-tuple-one-task kind of applications. anything more - a distributed-database environment will come into play. cheers, Vasu