1. Janusgraph-Users
  2. Topics

JanusGraph vs. Titan performance

"Lucie S." <svit...@...>
 

Hi,

I have a question regarding the performance of basic operations in JanusGraph. We use obsolete Titan in our project (with Persistit storage backend, so the Titan version is 0.4.4) and we wanted to compare it with JanusGraph in order to get rid of the deprecated database but to maintain (or better improve) our current performance. 

A brief report about our results is provided here: https://github.com/svitaluc/graphbench/tree/master/results. 
The source code of the benchmark can be found on the upper level here https://github.com/svitaluc/graphbench.

Based on the results I have a few questions:

- How is it possible that the import with JanusGraph is so slow in comparison with Titan? Is there any additional setting that we perhaps missed?

- Getting vertices with their neighbors/edges have tremendously worse performance than in the case of Titan. How is this possible? It seems to me like the model was changed and the edges are no longer stored in the same row because I do not think that there would be such a difference between Persistit and Cassandra. On the other hand, there would be no reason for the model change of course.

- Is the reasoning about cache behavior correct? Is there a part of cache that cannot be influenced? Maybe in the case of JanusGraph, this issue is negligible.

Thank you for any hint and response!

Cheers,
Lucie
Evgeniy Ignatiev <yevgeniy...@...>
 

Hello Lucie,

Seems that Persistit and Cassandra backends cannot be compared directly, as far as this one article benchmarked the Cassandra vs Persistit on TitanDB only - https://getmanta.com/blog/metadata-repository-benchmark-graph-database-titan/ - Persistit is way more performant, although isn't it a single machine only setup? I personally doubt that it will be easily comparable to Cassandra.

Also Cassandra 2.x is really a legacy, personally I would not consider it as a storage backend for production JanusGraph deployment (about 2 years ago, when I started to get acquainted with JanusGraph - we used Cassandra 3.x from the very beginning).

Maybe it is worth trying to use JanusGraph+ScyllaDB instead?

Best regards,
Evgenii Ignatev.

On 22.07.2020 12:10, Lucie S. wrote:

Hi,

I have a question regarding the performance of basic operations in JanusGraph. We use obsolete Titan in our project (with Persistit storage backend, so the Titan version is 0.4.4) and we wanted to compare it with JanusGraph in order to get rid of the deprecated database but to maintain (or better improve) our current performance. 

A brief report about our results is provided here: https://github.com/svitaluc/graphbench/tree/master/results. 
The source code of the benchmark can be found on the upper level here https://github.com/svitaluc/graphbench.

Based on the results I have a few questions:

- How is it possible that the import with JanusGraph is so slow in comparison with Titan? Is there any additional setting that we perhaps missed?

- Getting vertices with their neighbors/edges have tremendously worse performance than in the case of Titan. How is this possible? It seems to me like the model was changed and the edges are no longer stored in the same row because I do not think that there would be such a difference between Persistit and Cassandra. On the other hand, there would be no reason for the model change of course.

- Is the reasoning about cache behavior correct? Is there a part of cache that cannot be influenced? Maybe in the case of JanusGraph, this issue is negligible.

Thank you for any hint and response!

Cheers,
Lucie
--
You received this message because you are subscribed to the Google Groups "JanusGraph users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to janusgra...@....
To view this discussion on the web visit https://groups.google.com/d/msgid/janusgraph-users/27693e05-285f-4445-a715-59604ff5f995o%40googlegroups.com.

1 - 2 of 2