Re: JournalDiskSyncStrategy?

Karl Wettin wrote:

> What I do is to place a transient object graph of a given root
> object on a queue which is polled by a single thread which
> evaluate the delta which produce and execute transactions.
> Sometimes as many as 200 transaction from a single delta. 
>
> Usually I manage to empty this queue a every few seconds when
> not syncing on each transaction. It would be nice to sync at
> that point.
>
> Running FS.sync on each and every transaction produced from the
> delta of a transient root object graph takes up to 10 seconds
> on my slow HDD, usually 5-7, while not syncing takes 1-3
> milliseconds.

Oh, interesting. Prevayler tries to batch multiple transactions
into a single disk sync, but the current implementation is
thread-based: The number of transactions per batch is bounded by
the number of threads that are submitting transactions. You
probably don't want to create 200 threads just to submit those
200 transactions.

If the Prevayler interface were redesigned to be asynchronous,
you could submit all 200 transactions without waiting for them to
complete, and Prevayler could take care of batching them for you.
Unfortunately that would require some nontrivial redesign of
Prevayler's internals.

A less drastic change would be to add a single new method to the
Prevayler interface, e.g.:

void executeAll(Iterable<? extends Transaction<? super P>>)

The implementation could then ensure that all of the given
transactions get batched into a single disk sync.

For that matter, even without any changes to Prevayler, you could
combine your 200 individual transactions into a single composite
transaction with the same effect, e.g.:

public final class CompositeTransaction<P>
    implements Transaction<P>, Serializable {
  private final List<Transaction<P>> transactions;
  public CompositeTransaction(
      Collection<? extends Transaction<P>> transactions) {
    this.transactions = new ArrayList<>(transactions);
  }
  public void executeOn(P prevalentSystem, Date executionTime) {
    for (Transaction<P> transaction : transactions) {
      transaction.executeOn(prevalentSystem, executionTime);
    }
  }
}

Since you're not actually wanting to turn off disk sync, just
make it more efficient, one of these options might be preferable
to making the disk sync logic itself more configurable.

Cheers,
Justin

------------------------------------------------------------------------------
_______________________________________________
To unsubscribe go to the end of this page: http://lists.sourceforge.net/lists/listinfo/prevayler-discussion
_______________________________________________
"Databases in Memoriam" -- http://www.prevayler.org