3 ms·
How does this compare to Irmin[0][1][2], a similar system written by the guys working on Mirage? On first glance the one major difference I see is that Irmin su
by seliopou 12y ago
How does this compare to Irmin[0][1][2], a similar system written by the guys working on Mirage? On first glance the one major difference I see is that Irmin supports merges, while Pfs does not. Also, what exactly is specific to Docker here?
[0]: https://github.com/mirage/irmin https://github.com/mirage/irmin
[1]: http://openmirage.org/blog/introducing-irmin http://openmirage.org/blog/introducing-irmin
[2]: http://openmirage.org/blog/introducing-irmin-in-xenstore http://openmirage.org/blog/introducing-irmin-in-xenstore
- jdoliner 12y agoHi, I wasn't previously familiar with Irmin so I can't say this with certainty but it looks like the data models of pfs and Irmin are very similar. They are targeting slightly different use cases though so I expect some performance details to be different. Irmin bills itself as a database so it likely has much faster point access than pfs and more interesting indexing options. Pfs on the other hand is designed to handle much larger filesizes than Irmin. (Irmin doesn't list a limit but people don't normally put TB size rows in their databases.) The answer to your second question is also a big difference between pfs and Irmin. Although it's currently unimplemented pfs is designed to leverage docker for its MapReduce implementation. Rather than implementing a Java class to define a MapReduce job as one does in Hadoop, in pfs you create a docker image with a webserver inside to define you MapReduce job. This makes it trivial to include whatever dependencies you want in your distributed computations. Hope this clears some stuff up.