Tuesday, September 21, 2010

The power of connotations

Its funny how even intelligent engineers can be easily swayed by the connotations associated with a name.

Periodically, we get asked "why does Darkstar (now RedDwarf) use pessimistic concurrency, wouldn't optimsitic concurrency be better? Optimism is better then pessimism right?"

Well, not necessarily. Unbridled optimism can be a very bad thing. Optimism about the housing market created the economic situation we are all in now. But more to the point, the connotations of the words "optimistic" and "pessimistic" don't really say anything about how the software functions. As it turns out neither is "better" then the other. It depends on what you are trying to optimize, where and when.

To quote wikipedia:

"In the field of relational database management systems, optimistic concurrency control (OCC) is a concurrency control method that assumes that multiple transactions can complete without affecting each other, and that therefore transactions can proceed without locking the data resources that they affect. Before committing, each transaction verifies that no other transaction has modified its data. If the check reveals conflicting modifications, the committing transaction rolls back.[1]

However, if conflicts happen often, the cost of repeatedly restarting transactions hurts performance significantly; other concurrency control methods have better performance under these conditions."

So called pessimistic concurrency checks for conflicts when an object is first accessed for read or write. If there others out there using the object in a conflicting manner, it can cause the thread to pause while it waits for the other thread to finish its use of that object. In the event that there is such a contention, this is a good thing. It means you don't waste CPU doing calculations based on a 'stale' object and throwing all that work away at the end and having to start all over again. This reduces CPU load. While the blocked thread is waiting, other threads get to use the CPU. The end result is that more users can be processed in parallel with less total CPU usage.

However that check does come with some small cost. In an environment where you are accessing hundreds or thousands of objects in a thread this way, it can add up. Thats where so-called optimistic concurrency comes in. It doesn't take those costs but just acts like the object is always free. It keeps its own copy and, at the end, checks for consistency. If it finds a conflict, it dumps all its investment and starts over. (No government bailout applies.)

SO, which is better? It depends on your expected usage. If you expect a high volume of data accesses to data processing with very little contention, optimistic concurrency makes sense.

In Darkstar/RedDwarf however where we expect maybe a few dozen data access per thread, and where we expect our data processing (game logic) to be of significant cost, its better to be safe (and pessimistic) then sorry.

No comments: