Hello Taylor, Thank you very much of the clarification email. I am sure that it will be a vital reference point for many new and current users of Storm. I would like to suggest that a summary of your email becomes part of Apache Storm's documentation.
Kind Regards, Nikos On Thu, Jun 2, 2016 at 10:35 AM, P. Taylor Goetz <[email protected]> wrote: > There are a few things to keep in mind when evaluating Heron and Storm: > > First is performance. Twitter benchmarked Heron against a very old, > pre-Apache version of Storm (back when the transport layer was based on > 0mq), so their claims of performance improvements over Storm are likely > significantly overblown. There have been an enormous number of performance > improvements since then, and the Storm 1.0 release likely erases most of > the performance gain claimed by the Heron project. > > Second, despite their claims, Heron is not API compatible with the latest > release of Apache Storm. It may be somewhat compatible with the 0.9.x > series of releases, but 0.10.x is likely to have some compatibility issues > (I haven’t tested this out so I don’t know for sure), and it’s certainly > not compatible with 1.0. > > Finally, lets look at a few things that Storm has that Heron does not. Off > the top of my head I can think of: > > * End-to-end security (Kerberos, etc.), including secure integration with > other Apache Hadoop projects like ZooKeeper, HDFS, HBase, etc. > * Trident API (microbatching, exactly-once processing, etc.) > * Distributed Remote Procedure Calls (DRPC) > * Built-in windowing support > * State management (stateful bolts with automatic checkpointing) > * Distributed Cache API > * Kafka integration (though I believe this is coming) > * Integration with HDFS, Hive, HBase, Cassandra, Solr, Elastic Search, > Redis, MongoDB, JDBC, MQTT, and Azure Event Hubs. > * Scheduler framework independence (Heron requires Apache Mesos) > * Partial key groupings > * Declarative topology wiring (i.e. Flux) > > Is Heron a drop-in replacement for Storm? Probably not. > > -Taylor > > On Jun 2, 2016, at 9:27 AM, [email protected] wrote: > > Hi Marc, > > I had come across Heron a couple of weeks ago. It was indeed quite > interesting. Thanks for the hint. > > Regards > Leon > > > 1. Jun 2016 11:47 by [email protected]: > > > Maybe also take into account the new heron > > https://blog.twitter.com/2016/open-sourcing-twitter-heron > > > - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -. > F1 Outsourcing Development Sp. z o.o. > Poland > > t: +48 (0)124466845 > f: +48 (0)124466843 > e: [email protected] > > > -----Original Message----- > From: [email protected] [mailto:[email protected]] > Sent: woensdag 1 juni 2016 11:44 > To: User > Cc: [email protected] > Subject: Re: Storm unique strengths > > Hi Aaron, > > thank you very much for the link. I found it quite insightful. It is one > of the few benchmarks i have encountered where Storm comes out on top in > terms of latency, although the at-most once trade-off is quite harsh. > > Regards > Leon > > 31. May 2016 15:37 by [email protected]: > > > > Hi Leon, > > This isn’t an advocacy piece per se, but this analysis by several > member of the Storm community may be helpful. For a particular use case > you can compare performance and then assess whether the features, > user-friendliness, or API of a particular framework is worth switching > to. > > https://yahooeng.tumblr.com/post/135321837876/benchmarking-streamin > g-computation-engines-at > > > From: "[email protected]" <[email protected]> > Reply-To: "[email protected]" <[email protected]> > Date: Monday, May 30, 2016 at 3:28 AM > To: "[email protected]" <[email protected]> > Subject: Storm unique strengths > > > Hi Storm team, > > there are a lot of online comparisons between Storm and other Data > Stream Management Systems, yet few of them originate from Storm > committers/advocats. > I am trying to identify the aspects that Storm possesses, which > make it stand out among its direct competitors. Currently there is > significant competition from Apache Flink, although less so from Spark > due to its seconds latency restriction. > > From my experience Storm offers a unique support for DSLs, as well > as a very flexible concept of Spouts and Bolts. Other aspects however > seem to have been improved upon by Flink in greater part. > > Would you be able to direct me to resources that argue more towards > Storm's case? > > Thanks in advance. > > Leon > > > -- Nikos R. Katsipoulakis, Department of Computer Science University of Pittsburgh
