Compact Off-Heap Structures and Tuples In Java | Hacker News Reader

Compact Off-Heap Structures and Tuples In Java

mechanical-sympathy.blogspot.co.uk

30 pointspacoverdi12 comments

Funny, I was actually manipulating C structures from Java, and after hitting a nasty issue, went procrastinating on twitter where I stumbled upon this blog post :)

For those interested, there are less low-level ways to access fields in structures, for example Javolution http://javolution.org/target/site/apidocs/javolution/io/Stru...

I'm working on a very similar lib but where Structs are stateless, allowing to use them concurrently from multiple threads.

Just a quick question. For the flyweight idea in the article you get additional thread unsafeness even for immutable data in that you are changing where the flyweight points (which is how you want to do it to avoid creating and destroying a lot of flyweight objects). What is your idea to pool/control multiple flyweight accessors?

[Not sure whether the question is directed to the author or to my comment but I'll answer anyway :) ]

Basically, there are 3 notions

1) the buffer (byte[] or whatever) containing the data, immutable at least while the data is being accessed by a given thread

2) the Struct object: defines the structure of a message (field length, offset, arrays etc.)

3) the Message (aka the flyweight) that points to a given offset in buffer and is bound to a Struct instance

The buffer and the message are normally accessed by only one thread, but several threads can share Struct instances (i.e. the dictionary).

Each field (even deep in the field hierarchy) knows its offset relative to the beginning of the message, so values can be accessed like eg.

  int foo = struct.foo.getInt(msg);

The only complicated thing is when dealing with arrays (especially when nested in other arrays). The message does the bookkeeping necessary to safely access elements, ie. the mutable state is stored in the flyweight.

I like to have only one writer thread for any data. This can have its own flyweight. Other threads can be readers each having their own read only flyweight implementation.

Pretty much as expected. I'd be interested to see a comparison with object pooling (obviously not appropriate to the algorithm as written, but in a real system you're more likely to be streaming in data in chunks rather than putting it all in a big buffer and then reading through it), which lets you get a lot of the performance advantages of avoiding GC without completely abandoning Java's safety guarantees.

How would it work with object pooling if I wanted to query a large table of data? This is often needed in real big data applications.

For that kind of problem I'd probably be using hadoop, which does object pooling internally with the objects it passes into your mappers/reducers.

For a non-hadoop datasource you could do the same thing by hand: stream in the data from the table, turning it into objects from your pool and passing them through to your reducer function in small batches.

Interesting. It sounds like your issues are IO dominant since you do not mind the JVM startup cost from Hadoop for each query on each node. I'm more often looking at large data that is all memory resident which tends to drive the design this way. In finance queries need to have latencies way below sub-second which Hadoop cannot come close to satisfying. This is comparing batch to real-time analytics.

You're right that most of my big-data experience is batch work, and outside of finance. I guess I'm finding it hard to envision the kind of data where you'd want to work on the whole set, but that set's small enough to fit into memory - for real-time analytics wouldn't you be wanting to stream data and reduce it to the representation you want as it comes in?

In finance you may be re-evaluating a whole portfolio of assets, or doing a value at risk (VAR) calculation across everything. More often you want low-latency access to the entire dataset without going to disk. For this the entire data must be memory resident and compact.

That is both scary, and so very intriguingly interesting!

you can do this on heap with byte[] using Unsafe.arrayBaseOffset and getInt(Object, long), putInt(Object, long, int) and friends

Reply on news.ycombinator.com