Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 3 years ago.
Improve this question
What are the core architectural differences between these technologies?
Also, what use cases are generally more appropriate for each?
Update
Now that the question scope has been corrected, I might add something in this regard as well:
There are many comparisons between Apache Solr and ElasticSearch available, so I'll reference those I found most useful myself, i.e. covering the most important aspects:
Bob Yoplait already linked kimchy's answer to ElasticSearch, Sphinx, Lucene, Solr, Xapian. Which fits for which usage?, which summarizes the reasons why he went ahead and created ElasticSearch, which in his opinion provides a much superior distributed model and ease of use in comparison to Solr.
Ryan Sonnek's Realtime Search: Solr vs Elasticsearch provides an insightful analysis/comparison and explains why he switched from Solr to ElasticSeach, despite being a happy Solr user already - he summarizes this as follows:
Solr may be the weapon of choice when building standard search
applications, but Elasticsearch takes it to the next level with an
architecture for creating modern realtime search applications.
Percolation is an exciting and innovative feature that singlehandedly
blows Solr right out of the water. Elasticsearch is scalable, speedy
and a dream to integrate with. Adios Solr, it was nice knowing you. [emphasis mine]
The Wikipedia article on ElasticSearch quotes a comparison from the reputed German iX magazine, listing advantages and disadvantages, which pretty much summarize what has been said above already:
Advantages:
ElasticSearch is distributed. No separate project required. Replicas are near real-time too, which is called "Push replication".
ElasticSearch fully supports the near real-time search of Apache
Lucene.
Handling multitenancy is not a special configuration, where
with Solr a more advanced setup is necessary.
ElasticSearch introduces
the concept of the Gateway, which makes full backups easier.
Disadvantages:
Only one main developer [not applicable anymore according to the current elasticsearch GitHub organization, besides having a pretty active committer base in the first place]
No autowarming feature [not applicable anymore according to the new Index Warmup API]
Initial Answer
They are completely different technologies addressing completely different use cases, thus cannot be compared at all in any meaningful way:
Apache Solr - Apache Solr offers Lucene's capabilities in an easy to use, fast search server with additional features like faceting, scalability and much more
Amazon ElastiCache - Amazon ElastiCache is a web service that makes it easy to deploy, operate, and scale an in-memory cache in the cloud.
Please note that Amazon ElastiCache is protocol-compliant with Memcached, a widely adopted memory object caching system, so code, applications, and popular tools that you use today with existing Memcached environments will work seamlessly with the service (see Memcached for details).
[emphasis mine]
Maybe this has been confused with the following two related technologies one way or another:
ElasticSearch - It is an Open Source (Apache 2), Distributed, RESTful, Search Engine built on top of Apache Lucene.
Amazon CloudSearch - Amazon CloudSearch is a fully-managed search service in the cloud that allows customers to easily integrate fast and highly scalable search functionality into their applications.
The Solr and ElasticSearch offerings sound strikingly similar at first sight, and both use the same backend search engine, namely Apache Lucene.
While Solr is older, quite versatile and mature and widely used accordingly, ElasticSearch has been developed specifically to address Solr shortcomings with scalability requirements in modern cloud environments, which are hard(er) to address with Solr.
As such it would probably be most useful to compare ElasticSearch with the recently introduced Amazon CloudSearch (see the introductory post Start Searching in One Hour for Less Than $100 / Month), because both claim to cover the same use cases in principle.
I see some of the above answers are now a bit out of date. From my perspective, and I work with both Solr(Cloud and non-Cloud) and ElasticSearch on a daily basis, here are some interesting differences:
Community: Solr has a bigger, more mature user, dev, and contributor community. ES has a smaller, but active community of users and a growing community of contributors
Maturity: Solr is more mature, but ES has grown rapidly and I consider it stable
Performance: hard to judge. I/we have not done direct performance benchmarks. A person at LinkedIn did compare Solr vs. ES vs. Sensei once, but the initial results should be ignored because they used non-expert setup for both Solr and ES.
Design: People love Solr. The Java API is somewhat verbose, but people like how it's put together. Solr code is unfortunately not always very pretty. Also, ES has sharding, real-time replication, document and routing built-in. While some of this exists in Solr, too, it feels a bit like an after-thought.
Support: there are companies providing tech and consulting support for both Solr and ElasticSearch. I think the only company that provides support for both is Sematext (disclosure: I'm Sematext founder)
Scalability: both can be scaled to very large clusters. ES is easier to scale than pre-Solr 4.0 version of Solr, but with Solr 4.0 that's no longer the case.
For more thorough coverage of Solr vs. ElasticSearch topic have a look at https://sematext.com/blog/solr-vs-elasticsearch-part-1-overview/ . This is the first post in the series of posts from Sematext doing direct and neutral Solr vs. ElasticSearch comparison. Disclosure: I work at Sematext.
I see that a lot of folks here have answered this ElasticSearch vs Solr question in terms of features and functionality but I don't see much discussion here (or elsewhere) regarding how they compare in terms of performance.
That is why I decided to conduct my own investigation. I took an already coded heterogenous data source micro-service that already used Solr for term search. I switched out Solr for ElasticSearch then I ran both versions on AWS with an already coded load test application and captured the performance metrics for subsequent analysis.
Here is what I found. ElasticSearch had 13% higher throughput when it came to indexing documents but Solr was ten times faster. When it came to querying for documents, Solr had five times more throughput and was five times faster than ElasticSearch.
Since the long history of Apache Solr, I think one strength of the Solr is its ecosystem. There are many Solr plugins for different types of data and purposes.
Search platform in the following layers from bottom to top:
Data
Purpose: Represent various data types and sources
Document building
Purpose: Build document information for indexing
Indexing and searching
Purpose: Build and query a document index
Logic enhancement
Purpose: Additional logic for processing search queries and results
Search platform service
Purpose: Add additional functionalities of search engine core to provide a service platform.
UI application
Purpose: End-user search interface or applications
Reference article : Enterprise search
I have been working on both solr and elastic search for .Net applications.
The major difference what i have faced is
Elastic search :
More code and less configuration, however there are api's to change
but still is a code change
for complex types, type within types i.e nested types(wasn't able to achieve in solr)
Solr :
less code and more configuration and hence less maintenance
for grouping results during querying(lots of work to achieve in
elastic search in short no straight way)
I have created a table of major differences between elasticsearch and Solr and splunk, you can use it as 2016 update:
While all of the above links have merit, and have benefited me greatly in the past, as a linguist "exposed" to various Lucene search engines for the last 15 years, I have to say that elastic-search development is very fast in Python. That being said, some of the code felt non-intuitive to me. So, I reached out to one component of the ELK stack, Kibana, from an open source perspective, and found that I could generate the somewhat cryptic code of elasticsearch very easily in Kibana. Also, I could pull Chrome Sense es queries into Kibana as well. If you use Kibana to evaluate es, it will further speed up your evaluation. What took hours to run on other platforms was up and running in JSON in Sense on top of elasticsearch (RESTful interface) in a few minutes at worst (largest data sets); in seconds at best. The documentation for elasticsearch, while 700+ pages, didn't answer questions I had that normally would be resolved in SOLR or other Lucene documentation, which obviously took more time to analyze. Also, you may want to take a look at Aggregates in elastic-search, which have taken Faceting to a new level.
Bigger picture: if you're doing data science, text analytics, or computational linguistics, elasticsearch has some ranking algorithms that seem to innovate well in the information retrieval area. If you're using any TF/IDF algorithms, Text Frequency/Inverse Document Frequency, elasticsearch extends this 1960's algorithm to a new level, even using BM25, Best Match 25, and other Relevancy Ranking algorithms. So, if you are scoring or ranking words, phrases or sentences, elasticsearch does this scoring on the fly, without the large overhead of other data analytics approaches that take hours--another elasticsearch time savings.
With es, combining some of the strengths of bucketing from aggregations with the real-time JSON data relevancy scoring and ranking, you could find a winning combination, depending on either your agile (stories) or architectural(use cases) approach.
Note: did see a similar discussion on aggregations above, but not on aggregations and relevancy scoring--my apology for any overlap.
Disclosure: I don't work for elastic and won't be able to benefit in the near future from their excellent work due to a different architecural path, unless I do some charity work with elasticsearch, which wouldn't be a bad idea
If you are already using SOLR, remain stick to it. If you are starting up, go for Elastic search.
Maximum major issues have been fixed in SOLR and it is quite mature.
Imagine the use case:
A lot(100+) of small(10Mb-100Mb, 1000-100000 documents) search indexes.
They are using by a lot of applications (microservices)
Each application can use more than one index
Small by size index, yes. But huge load(hundreds search-requests per second) and requests are complex (multiple aggregations, conditions and so on)
Downtimes are not allowed
All of that is working years long, and constantly growing.
Idea to have individual ES instance per each index - is huge overhead in this case.
Based on my experience, this kind of use case is very complex to support with Elasticsearch.
Why?
FIRST.
The major problem is fundamental back compatibility disregard.
Breaking changes are so cool!
(Note: imagine SQL-server which require you to do small change in all your SQL-statements, when upgraded... can't imagine it. But for ES it's normal)
Deprecations which will dropped in next major release are so sexy!
(Note: you know, Java contain some deprecations, which 20+ years old, but still working in actual Java version...)
And not only that, sometimes you even have something which nowhere documented (personally came across only once but... )
So. If you want to upgrade ES (because you need new features for some app or you want to get bug fixes) - you are in hell. Especially if it is about major version upgrade.
Client API will not back compatible. Index settings will not back compatible.
And upgrade all app/services same moment with ES upgrade is not realistic.
But you must do it time to time. No other way.
Existing indexes is automatically upgraded? - Yes. But it not help you when you will need to change some old-index settings.
To live with that, you need constantly invest a lot of power in ... forward compatibility of you apps/services with future releases of ES.
Or you need to build(and anyway constantly support) some kind of middleware between you app/services and ES, which provide you back compatible client API.
(And, you can't use Transport Client (because it required jar upgrade for every minor version ES upgrade), and this fact do not make your life easier)
Is it looks simple & cheap? No, it's not. Far from it.
Continuous maintenance of complex infrastructure which based on ES, is way to expensive in all possible senses.
SECOND.
Simple API ? Well... no really.
When you is really using complex conditions and aggregations.... JSON-request with 5 nested levels is whatever, but not simple.
Unfortunately, I have no experience with SOLR, can't say anything about it.
But Sphinxsearch is much better it this scenario, becasue of totally back compatible SphinxQL.
Note:
Sphinxsearch/Manticore are indeed interesting. It's not Lucine based, and as result seriously different. Contain several unique features from the box which ES do not have and crazy fast with small/middle size indexes.
I have use Elasticsearch for 3 years and Solr for about a month, I feel elasticsearch cluster is quite easy to install as compared to Solr installation. Elasticsearch has a pool of help documents with great explanation. One of the use case I was stuck up with Histogram Aggregation which was available in ES however not found in Solr.
Add an nested document in solr very complex and nested data search also very complex. but Elastic Search easy to add nested document and search
I only use Elastic-search. Since I found solr is very hard to start.
Elastic-search's features:
Easy to start, very few setting. Even a newbie can setup a cluster step by step.
Simple Restful API which using NoSQL query. And many language libraries for easy accessing.
Good document, you can read the book: . There is a web version on official website.
Related
In terms of
scalability,
performance,
maintenance,
Ease of use / Learning curve
cost,
In order of significance but wouldn't mind a general answer as I appreciate I m probably asking for too much :)
Thanks
EDIT: I m looking for a database that will serve as the single authoritative data store and I need all attributes of the documents stored to be indexed for various business reasons. Therefore I know that other solutions won't do what I m looking for.
tl;dr; If you are using JavaScript and building browser apps, node.js and DocumentDB are a match made in heaven. If you are using .NET and/or other Azure services, then DocumentDB is favored. If you are using other AWS services, then SimpleDB might be better.
I know that questions like this are not ideal for Stack Overflow, but I often see value in answers like this and my most popular answer on SO is essentially informed opinion backed by evidence. I have not used SimpleDB but I looked into it before deciding on DocumentDB. I rejected it pretty quickly... although I did give AWS Lambda a serious look before deciding on DocumentDB. So:
scalability. DocumentDB has a very straight forward and explicit scaling model -- add more collections if you need either more space or more operations per second. SimpleDB's scaling model is similar except less straight forward since you add domains which are overloaded to both provide type separation (think tables) and scalability. You can scale either to whatever you need.
performance. Since I never built anything on it, I can't say anything about SimpleDB's performance. However, I've been very impressed with the performance of DocumentDB. Latency is less than 10ms for simple id-based reads and I get impressive latency and throughput for queries. The DocumentDB implementation of our current app returns complex n-dimensional aggregations (done in stored procedures on DocumentDB using documentdb-lumenize) in 1/4 the time of the functionally-equivalent MongoDB/node.js implementation. You'd have to do your own performance testing on your actuall application to have a definitive answer here.
maintenance. Both are much more hands off than traditional data stores. There just aren't that many knobs to turn maintaining either of them. SimpleDB geographically distributes your data by default. You'd have to do the equivalent manually in DocumentDB. Possible, but harder. DocumentDB has good import/export tools and their backup solution is about to be significantly upgraded.
ease of use / learning curve. If you are JavaScript programmer, than DocumentDB has a lot to recommend it. DocumentDB uses JSON natively. SimpleDB uses XML. DocumentDB has ACID-enabling stored procedures written in JavaScript. You'd need to combine SimpleDB with something else (Lambda maybe, but the XML/JavaScript mismatch would make this less than ideal) to get the equivalent. Both allow use to use SQL but DocumentDB also allows for JavaScript native queries.
There is one huge mindset hurdle that you will have to get over in order to be successful with DocumentDB. Despite the fact that they both scale by adding more domains/collections, SimpleDB domains are closer conceptually to tables. The word choice of "collection" by the DocumentDB team is unfortunate since they are more akin to partitions and should not be thought of as tables. The hard part is getting used to the idea that you store all of your different data types in the same collection. Once you get over that, I find DocumentDB's approach refreshing and incredibly flexible. I can efficiently model inheritance and type-mixins. Collections nay partitions have one purpose -- scalability. Domains are used for both scalability and data type separation which is actually harder in practice.
cost. Not much to say here. Both allow you to scale your cost gradually. For really small implementations, DocumentDB is probably more expensive since the smallest unit of usage is a single collection which is $25/month minimum. You'd have to do your own modeling/what-if analysis to determine which would be less expensive for you. Note, that Azure is being every aggressive in general and even pushing AWS to lower prices in some cases. My gut is that they would be roughly equal in cost for most applications.
Other thoughts:
You wrote, "I need all attributes of the documents stored to be indexed". One really nice feature of DocumentDB is that you can specify the size of your indexes By default, every field is indexed into a 3-byte per field hash index, which is highly space efficient. I do not know if SimpleDB has the equivalent.
This is a bit like comparing apples to oranges. I consider DocumentDB to be like MongoDB or CouchDB in it's data model and VoltDB in its use execution model (although VoltBD sprocs are written in Java). SimpleDB feels more like a simple XML object store. If you already have a big XML mindset, then it might be easier, but I think there are more folks using JSON today than XML.
Writing ACID-enabling stored procedures in JavaScript is a killer feature that only DocumentDB has. Some say the days of stored procedures are over; that you should put all such logic in your application server layer. If you implementing a simple CRUD API, that may be, but almost every application requires some sort of transaction where more than one row is changed at a time. This is mind bogglingly hard to do correctly without transaction support in your data store. Even if you do implement the equivalent of transactions with your NoSQL database, the overhead of the implementation eats away any development/performance/scalability advantages that you got by choosing NoSQL rather than SQL.
DocumentDB's user defined functions and triggers (also written in JavaScript) might be useful, although I believe the trigger implementation is crippled at this moment in time and I haven't found a use for UDFs myself yet.
DocumentDB has built-in attachment support. You need to integrate manually with S3 for the equivalent on AWS.
DocumentDB has geo indexing and operators.
SimpleDB's 1K per document limit is a serious limitation. This tells me that it's designed mostly for logging or as an index to S3 and not a full-fledged document store. The limit for DocumentDB is 512K.
If comparison to SimpleDB is like apples to oranges, then comparison to ElasticSearch is like apples to fire engines. My impression of ElasticSearch is that it's all about full-text searching and analytics. I don't think it's space/execution/api efficient to serve as a primary transactional store. Built on Lucene, it was not designed to have the reliability/durability to be your primary store. Further, even when hosted, it's more of an IaaS offering, wherease DocumentDB and SimpleDB are true PaaS offerings. The maintenance will be much higher with ElasticSearch.
Is there anybody out there using Node-Neo4j-embedded in production mode ?
What kind of limits are expectable ?
Because I think this module is pushing the Cypher queries directly to the node-java module, what uses them directly with Neo4j java libs, I belief there shouldn't be any limits.
I feel it is dangerous to decide to use a lib what isn't maintained for about 2 years (see: github) - and it shouldn't be on Neo4j docs if it isn't maintained (see: README.md dead link about API-Docs).
It looks like there could be a new trend to power up node.js support like first citizen languages by other distributor(s) for (in_memory) graph databases. Maybe Neo4j also should review this and the unmaintained node module (like OrentDB did). The trend had bin initiated by a benchmark-battle between ArangoDB and OrientDB.
I would love to see an Node-Neo4j-embedded benchmark answer to the open source benchmark of ArangoDB - done by professional Neo4j people like OrientDB people had done. But note: They hadn't been fair enough (read the last lines about enabling query caches...).
Or it has to be a new benchmark focused on most possible first citizens-like access by NodeJS. There are three possible solutions to test. I am not experienced enough to do such a test what would be really acceptable. But I would like to help by verifying this.
Please support this call for action with comments and (several types of) answers. A better (native like access) and wider range of supporting in_memory and graph solutions would help the node community very much. A new benchmark would force innovation
Short note about ArangoDBs benchmark: They've tested the REST-APIs. But if you think about performance, you don't like to use a REST-API - you like to use direct library access.
#editors: you are welcome
We (ArangoDB) think that the scalability of embedded databases is to limited. It also limits the number of databases which you may want to compare. Users prefer to implement their solutions in their Application stack of choice, so you would limit the number of people potentially interested in your comparison.
The better way of doing this is to compare the officialy supported interface of the database vendor into the client stack that is commonly supported amongst all players in the field. This is why we have chosen nodejs.
There is enough chatter about benchmarks and how to compare them on stackoverflow, so in doubt, start out to create a usecase and implement code for it, present your results in a reproducable way and request that for comment, instead of demanding others to do this for you.
I would like to build a search engine for my website so I can quickly find relevant content. I've done quite a few google searches, discovered ElasticSearch and Solr (which both sit on top of Lucene), and whoosh (python-based).
But are all of these search engines just building an "inverted-index" on top of the data? What are some other algorithmic approaches for getting higher quality searches?
I was intrigued by this blog post using collaborative filtering on top of Solr, which returns related search queries:
http://www.opensourceconnections.com/2013/08/25/semantic-search-with-solr-and-python-numpy/
Are there other common techniques that I should be aware of? Are there other libraries sitting on top of ElasticSearch/Solr that I could just plug into, and use "out-of-the-box"?
Any links or tips would be greatly appreciated!
You haven't mentioned what tech stack you are working on.
If you use Ruby on Rails, I would recommend Tire, which is a gem that gives a DSL wrapper over ElasticSearch. Essentially, it allows you to index your data in Elasticsearch.
For Rails, Sunspot is a very popular gem that people use to interface with Solr.
For .NET - SolrNET is a great Solr client.
Other part of your question (around implementing a good search engine) is too broad - I would recommend reading a good book such as Lucene in Action to get a feel of what Solr/Elasticsearch could do.
I do have a few notes that I wrote a while back, you can read about some of my experience in search here.
Edit:
Since you work on python, I would recommend Haystack, although it is specific to Django. It is very versatile for our needs. However, if you are not using django, I can think of solrpy as a Solr client. Haystack works with both Solr and Elasticsearch.
i suggest you to learn Solr API, cause it was developed since 4 5 years so you can find lots of plug-ins like related search API in Solr, But in elastic search it is very easy to configure however it is very young engine so needs to be developed more.
Pyes is a well-documented Python client for Elasticsearch.
Also, this Youtube video provides a good overview of using Elasticsearch with Python.
I suggest you to use Google Custom Search Engine.
Here have a look.
https://www.google.com/cse/all
We have developed several search engines both on Solr and Elastic. Solr used to be the best as it provided most of the tools needed to admin and debug your indexes. Right now Elastic offers the same features as Solr either natively or via plugins. Plus it is easier to configure in high performance/high availability scenarios (easy to shard or cluster).
Your technology stack is irrelevant. Both Solr and Elastic have clients nearly for every language, plus you can access both via plain HTTP:
That said, each search engine applies to a problem domain. Tunning Elastic or Solr to retrieve relevant results is a bit of an art with some trial and error.
You will have to define analyzers for each field you'll search on and according to your search patterns and the kind of results you will be expecting.
Eventually, to create search engines with a single input that search across disparate attributes of a document type, may need the use of DisMax queries where you can boost results depending on the matching of the search terms to specific document fields.
To summarize: go for Elastic, and get some plugins or frontends. Two suggestions:
Inquisitor: for testing your analyzers
Elastic Head: for administration purposes
Two clients of mine are evaluating setting up a search server, either Solr or ElasticSearch. We're wondering what programming languages (if any) and development environments are necessary to get the search servers running. Can it be done by people mostly familiar with front end technologies (HTML/CSS/JavaScript) or is more serious coding skill needed (e.g. understanding of multithreading/ advanced debugging/ other "pro-level" concepts)?
If only light programming skills are needed I'm playing with the thought of suggesting to set it up myself. I have very little Java knowledge but have basic understanding of C, ActionScript, Pascal and even Simula in addition to aforementioned front end technologies. I know basic search architecture from my time in FAST (an enterprise search vendor).
Best, Bjørn
Bit of a broad question but i'll try to give it a shot:
You don't need any programming language in particular. They're both stand alone servers which have API's which are addressable from any programming language.
ElasticSearch has a really nice API that's JSON/REST based.
SOLR's API is a lot more clunky, but also supports XML.
(If I have a choice I tend to go for ElasticSearch, unless there's a really specialized feature I need that's only in SOLR).
Getting up and running doesn't really require any knowledge of any programming language in particular.
The only time you NEED java is when you decide you end up needing custom plugins to SOLR/ElasticSearch itself.
You don't need any specific IDE's beyond those matching your programming language of choice.
When trying to figure out what's going on inside my elasitc search server I do like elastic search HEAD:
http://mobz.github.io/elasticsearch-head/
Hope this helps.
As pointed out already, this is quite a broad question, most likely get closed. But I'll give it a go too.
Both ElasticSearch and Solr are quite easy to get started with. They come as a zip/tar.gz archive that you can extract.
Both require JVM, so you need Java setup.
Once setup, playing with either is quite easy, you do not need any advanced programming skills to play around with it. Solr comes with an Admin UI page, that allows you to execute queries.
Elastic Search has clients as #Constantijin has pointed out. Elastic-head is an excellent choice.
You will need quite a detailed understanding of the Lucene ecosystem, its architecture, plugins etc. Given you have an understanding of another Search Engine, the concepts around indexing and text processing should be easy enough for you.
If you want to write something more advanced than the Admin UI, and you can use Javascript.
You can use AjaxSolr for making ajax requests to your Solr instance
For ElasticSearch, you can try using Elastic.js.
Elasticsearch is an open-source search engine built on top of Apache Lucene™, a full-text search-engine library. Lucene is arguably the most advanced, high-performance, and fully featured search engine library in existence today—both open source and proprietary.
However, Elasticsearch is much more than just Lucene and much more than “just” full-text search. It can also be described as follows:
A distributed real-time document store where every field is indexed and searchable
A distributed search engine with real-time analytics
Capable of scaling to hundreds of servers and petabytes of structured and
unstructured data
I would like add more details regarding how to used ElasticSearch in php language check out - http://www.multidots.com/what-is-elasticsearch
[How to integrate ElasticSearch with PHP?][1]
By using curl, you can use ElasticSearch with your favorite programming language. Here is the example of simple curl request with ElasticSearch.
- PHP Sample Script:
You can find PHP client api on github:
[https://github.com/elastic/elasticsearch-php][2]
Check out Best Article on Elasticsearch - http://www.multidots.com/what-is-elasticsearch
Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 8 years ago.
Improve this question
I am looking for an eventually consistent data store and it looks like it may be coming down to Riak or Cassandra. Has anyone got expereinces of a view on this?
As you probably know, they are both architecturally strongly influenced by Dynamo (eventually consistent, no single points of failure, etc). Both also go beyond Dynamo in providing a "richer than pure K/V" data model -- in Cassandra's case, providing a Bigtable-like ColumnFamily mode, in Riak's, a Document-oriented one. I have seen sane people choose both.
I believe points that favor Cassandra include
speed
support for clusters spanning multiple data centers
big names using it (digg, twitter, facebook, webex, ... -- http://n2.nabble.com/Cassandra-users-survey-tp4040068p4040393.html)
Points that favor Riak include
map/reduce support out of the box
/Cassandra dev, fwiw
Riak is used by
Mozilla Foundation
Ask.com sponsored listings
Comcast
Citigroup
Bet365
I think they both pass the test of credible reference customers/users.
Cassandra seems more mature, and is currently doing better in benchmarks. Riak seems easier to add a node to as your cluster grows.
For completeness: A good (probably biased) comparison between the two can be found at http://docs.basho.com/riak/1.3.2/references/appendices/comparisons/Riak-Compared-to-Cassandra/
Use and download are different. Best to get references.
Perhaps a private conversation could be had where Riak references in these companies could be shared? Not sure how to get such with Cassandra, but there is a community of companies that support Cassandra that seem like a good place to start. As these probably have community participants in Cassandra development, it may be a REALLY reasonable place to start.
I would like to hear Riak's answer to recent and large deployments where customers are happy.
I also would like to see the roadmap for each product. Cassandra is a bit easier to track (http://wiki.apache.org/cassandra/) than Riak in my view as Cassandra's wiki discusses limitations and things that are probably going to change going forward, but neither outline futures well. I could understand that of an open source community ... perhaps ... but I cannot for a product for which I must pay.
I also would suggest research of Cloudant, which has what appears to be a very nice layering of capabilities. It also looks like it is bringing to bear the capabilities elsewhere in Apache land. CouchDB is the Apache platform on which Cloudant is based. BUT the indexing with Lucene seems but the tip of the iceberg when it comes to where Cloudant could go. Creating and managing an index is a very systematic process, a kind of data pipeline, that could be scripted using other Apache community assets. AND capabilities like NLP also could be added through Lucene indirectly, or maybe directly into what is persisted.
It would be nice to see a proposed Cloudant roadmap, especially since the team could mine the riches of the Apache community and integrate such into Cloudant. Such probably exists as there is an operational component to the Cloudant revenue model that will require it, if for no other reason.
Another area of interest ... Cloudant's pricing model ... it is clear their revenue model is not based on software, but around service. That is quite attractive, and it seems consistent with the ecosystem surrounding Cassandra too. I don't know if the Basho folks have won over enough of the nosql community as yet ... don't see such from any buzz around their web site or product.
I like this Cloudant web page (https://cloudant.com/the-data-layer/). I was surprised to see the embedded Erlang capability ... I did not know CouchDB was written in Erlang as this seems unusual to me in the Apache community (my ignorance); CouchDB appears to be older than other nosql products I know (now) to be written in Erlang. Whatever their strategy, they at least count Amazon EC2 and Microsoft Azure as hosting partners, indicating an appreciation of Microsoft and !Microsoft worlds - all very important if properly recognizing the middleware value potential (beyond cache or hash table applications) that these types of data stores could have.
Finally, while I don't know the board well, Andy Palmer's guidance looks like it will be valuable. He can bring some guidance vis-a-vis structured data (through VoltDB) to a world that rightly or wrongly may be unfairly branded as KVP hash tables of unstructured data. The need for structure and ecosystem surrounding nosql "databases" is being recognized ... witness Google's efforts with Spanner ... KVP/little structure/need for search-ability motivated Google's investment in the Spanner space. While we all may not need something like Spanner, we probably do need an improving and robust "enterprise" management and interoperability capability in these nosql databases to make it reasonable to incorporate them into modern cloud architectures. The needed structure can come from ease of interoperability and functional richness. It can also come from new capabilities that support conversion of unstructured data to structured data (e.g. indexes, use of NLP to create structured and parsed renderings of things inside of a KVP blob, and plenty of other things that, if put into a roadmap and published, could entice and grow a user base). Cloudant looks like it has a good chance of success ... I will take a closer look at it ...
And look what I found about CouchDB ...
CouchDB comes with a suite of features, such as on-the-fly document transformation and real-time change notifications, that makes web app development a breeze. It even comes with an easy to use web administration console. You guessed it, served up directly out of CouchDB! We care a lot about distributed scaling. CouchDB is highly available and partition tolerant, but is also eventually consistent. And we care a lot about your data. CouchDB has a fault-tolerant storage engine that puts the safety of your data first.