- From: Anselm Baird-Smith <anselmb@eng.sun.com>
- Date: Wed, 30 Jun 1999 09:03:03 -0700 (PDT)
- To: kabir@our.st
- Cc: jigsaw <www-jigsaw@w3.org>
This package really is a java port of gdbm (same algorithms, etc). kabir@our.st writes: > Question regarding jdbm class: > > Is this implementation thread safe? I've noticed that bucket lookups are synchronized, but I am not clear whether this guarantees thread-safety (or rather data integrity). It should be, the jigsaw is MT and uses it quite a bit ;-) Some methods are *intentionnally* not synchronized (for performance reasons) > I have also noticed some of the commented caveats in the jdbmBucket.java -- for example in jdbmBucket.add() regarding bucket splits -- but unfortunately I am not exactly clear on what actions would create such conditions. Can't remember what the comment says ;-) > Please note that I intend to use only the exposed jdbm interface and would not directly utilize (or subclass) the support classes; and that the objects that I am writing will guarantee that at any given time, only one thread will take destructive (w/u/d) action for a given key. That's the way it should be used > I would also greatly appreciate any information regarding usage and performance considerations (threading, file size caps, performance tweaks, etc.) for this class. (I am aware of the key Enumeration issue.) I;'ve probably done some of that testing, but nothing was ever written about it. In general it should behave similarly to gdbm Anselm.
Received on Wednesday, 30 June 1999 12:06:25 UTC