Following the acquisition of Aurelius by DataStax, and since Titan 1.0.0 was released in September 2015 and received very few commits since then, I wonder whether it is safe to use TitanDB in production. Can anyone please give some insight?
We are running a Titan graph, on top of HBase, with about 30TBs of data, and a lot of features are missing.
For example, a must-have is the ability to perform OLAP operations on the graph, such as deleting redundant vertices using Spark.
Although it seems like Tinkerpop is going that way using its SparkGraphComputer, it does not work well - the implementations of reading data from HBase using Hadoop InputFormat is buggy, and a lot of scenarios are not handled (for example, a vertex that is connected to itself in a loop makes the code raise an exception and terminate). In addition, the efficiency of the sections that parse vertices from raw data is just bad - a lot of buffer allocations that are redundant and make everything slow.
If you are planning on a big graph for the long term, I don't think Titan is suitable - unless you are going to contibute code by your own.
Apparently TitanDB has been forked and now is JanusGraph, here is an article about it.
[EDITED 28 April 2017 to note that Titan is indeed dead; JanusGraph is a fork of Titan and very much healthy]
I know of several groups using Titan in production, including people at IBM and Amazon. Is it safe? Absolutely, if you have staff that can support it. (Shameless plug: my company will help you with that.) Will Titan's development continue? TinkerPop development is still moving very fast, though contributions to core Titan are now mostly coming from non-Aurelius programmers.
I'd argue if you're a fan of Titan's approach because of your data sizes or velocities, you should be looking at moving to DataStax's new graph database when it's released sometime this year. Work on that proceeds at a feverish pace from what I can tell talking to the ex-Aurelius guys at various conferences. If you do your work in Titan 1.0 (or 1.1, which isn't far) using TinkerPop 3, then it should be a straightforward transition to DSE Graph. This is what we are suggesting to our clients who need to manage very large data loads.
It is NOT safe to use TitanDB at the moment because main maintainers dropped it and the last commit was over a year ago. The company was acquired by Datastax and the team was shifted to work on their DSE Graph. Though, few fellows run Titan 1.0 on production.
A natural alternative is JanusGraph which is essentially a well maintained re-branded fork of the unofficially released Titan v1.1. Check their Github repo.
© 2022 - 2024 — McMap. All rights reserved.