I'm not posting notices of new snapshots on this page any longer. They can be found on the download page instead.

There's a new snapshot available on the download page. It's mostly bug fixes from the previous one, including a rather serious bug in the DataManager stage that could lead to performance degredation over time. Also, the beginnings of a tutorial on bamboo.lss.ASyncCore can now be found in bamboo/doc/async-tutorial. It's heavily modelled after the libasync tutorial from MIT, but translated into Java.

Alas, I've had to roll back to the older, C-based version of Berkeley DB to get good reliability in the put/get layer. That said, it is very reliable now. See the latest snapshot on the download page for more details.

The latest snapshot on the download page uses Berkeley DB Java Edition, so Bamboo is now 100% pure Java.

Thanks to Marcel Dischinger, there is now a tutorial available on using Bamboo. Please take a look, and send any questions or comments his way.

Here's a picture of the Vivaldi implementation on PlanetLab. Note that all of the major geographical areas are reasonably well differentiated.

Two big developments today. First, I added an event-driven client-side interface to the OpenHash gateway in bamboo.dht.GatewayClient, along with an event-driven version of bamboo.dht.GatewayTest in bamboo.dht.GatewayClientTest. Then I implemented in the ReDiR algorithm on top of it; this algorithm allows a client to consistently hash across a subset of the nodes in the DHT, as opposed to all of them. It can also be used to hash across nodes that are not even in the DHT. For example, you could use this to load balance requests to a small number of nodes running an application server, without those nodes having to run a DHT of their own. See our IPTPS paper for more details.

The second big addition to the code is an implementation of the Vivaldi algorithm, which computes virtual network coordinates for each DHT node such that the Euclidean distance between two nodes' virtual coordinates is proportional to the network latency between them. As such, the coordinates are useful for things like server selection. See this HotNets paper for more information.

Anyway, both of these additions are available in today's CVS snapshot. They're both pretty alpha code at this point, but they're also both actively being worked on, so please feel free to give them a shot in your application. I'm especially interested in anyone using ReDiR, so please drop me a line if it sounds interesting to you.

I just got a Powerbook, so I went ahead and compiled the Berkeley DB Java libraries for OS X. They're in bamboo/lib/darwin as of today's CVS snapshot, and all of the tests work just fine with them. This addition means that Bamboo should now work out of the box on Linux, FreeBSD, and OS X. I don't have any Windows computers, but if someone will send me the relevant recompiled libraries, I'll add them, too.

Also, the bamboo.db.StorageManager class was leaking Berkeley DB cursors. Be sure to update to today's snapshot to get the fix.