Add a comparison to similar NoSQL databases to the docs and readme file

[thedrow]

It's much easier to understand how a new tool works if you can read how it behaves differently from other tools we've already used. I'd like to see a comparison with Redis for example.

[tammet]

10/27/2013 3:06 AM, Omer Katz kirjutas:

It's much easier to understand how a new tool works if you can read how it behaves differently from other tools we've already used. I'd like to see a comparison with Redis for example.

A good idea. Will attempt to add some brief comparisons with, say, redis, mongodb, Tokyo cabinet and neo4j in near future. However, this might easily open a can of worms with people being unhappy about the way we present the differences, i.e. like one guy wants to stress features A and B of redis, another features C and D, etc etc.

Very briefly said, redis is an advanced distributed key-value cache server like memcached, whitedb is a graph/N-tuples database shared memory library, like sqlite without SQL but with record pointers and operating in shared memory, not files.

The main differences with Redis are imho these:

• redis is a key-value base (search a value assigned to a given key) whitedb is an N-tuple base (search from any of the fields in a tuple for a whole tuple of values)
• in other words, redis allows to search only by the exact unique key in the hash, whitedb allows to search also for a set of equal or smaller/greater values from a T-tree index, from any field, like conventional relational databases.
• redis is a database server to which clients connect over tcp/ip. whitedb is a library with no running server process: processes using the whitedb library access data stored in a shared memory segment.
• for most operations whitedb is faster than redis: no time is spent on client/server connections.
• redis has a machinery for partioning data over several redis instances (machines), whitedb has no such mechanism: all the data is stored in one machine.
• redis has transactions (atomic sequences of updates, yet no rollbacks), whitedb has no transactions and relies on read/write locks only (different locking strategies available).
• the persistance mechanisms (storing to disk) are similar for redis and whitedb: both offer database dumps/snapshots and logging on demand.

Tanel

[thedrow]

Can you add to each comparison what are the advantages and disadvantages of replacing X with WhiteDB for the typical use case of X? Suggesting hybrid approaches where appropriate or recommending to use X or WhiteDB entirely for that use case will certainly contribute to the adoption of WhiteDB.

Also, benchmarks for using WhiteDB over X for the typical use case of X can be useful.

It is also important to highlight what differences are by design and what might be implemented in the feature (with links to issues if applicable). For example is the lack of partitioning by design? If so, why? If we need partitioning should we prefer Redis or should that feature be emulated?

You get the idea right?

Overall you have a great product here but I can't really tell if I should be using it instead of or in addition to another tool.

[thedrow]

Regarding the can of worms remark, the documentation should include all differences. If a user wants to highlight another use case it should be added to the documentation/wiki. This is open source right? Pull requests are welcome :)

[tammet]

10/27/2013 2:19 PM, Omer Katz kirjutas:

Can you add to each comparison what are the advantages and disadvantages of replacing X with WhiteDB for the typical use case of X? Suggesting hybrid approaches where appropriate or recommending to use X or WhiteDB entirely for that use case will certainly contribute to the adoption of WhiteDB.

Also, benchmarks for using WhiteDB over X for the typical use case of X can be useful.

It is also important to highlight what differences are by design and what might be implemented in the feature (with links to issues if applicable). For example is the lack of partitioning by design? If so, why? If we need partitioning should we prefer Redis or should that feature be emulated?

You get the idea right?

Overall you have a great product here but I can't really tell if I should be using it instead of or in addition to another tool.

Sure, just that writing down all these comparisons, suggestions and benchmarks is a massive amount of work which we would love to do but most likely just cannot in such detail :). Now, it would be a great project for a student: I'll probably really propose it to my students as a possible thesis subject.

Tanel

[tammet]

10/27/2013 2:25 PM, Omer Katz kirjutas:

Regarding the can of worms remark, the documentation should include all differences. If a user wants to highlight another use case it should be added to the documentation/wiki. This is open source right? Pull requests are welcome :)

Exactly. You are most welcome to add to the documentation!

What I wanted to stress is that our small team cannot realistically "include all differences": there are far too many database systems, far too many differences and aspects of these differences for us to cover even a small percentage in sufficient detail :)

Tanel

[thedrow]

I will if I'll end up using WhiteDB. Right now I just need a basic comparison or a use case so I'll know when should I introduce WhiteDB. Instead of memcached? Redis? Neo4J? In scenario X where all of those options are not optimal?

[tammet]

10/27/2013 2:45 PM, Omer Katz kirjutas:

I will if I'll end up using WhiteDB. Right now I just need a basic comparison or a use case so I'll know when should I introduce WhiteDB. Instead of memcached? Redis? Neo4J? In scenario X where all of those options are not optimal?

Please be more concrete. Which of the ones above you actually use and for what. I'd say that in most cases where you already use one of these systems and are satisfied, just go on and continue using what works for you. You should probably not use whitedb if you cannot clearly see the benefits for your concrete project yourself.

Tanel

[thedrow]

I mean, it's unclear what kind of problem WhiteDB is trying to solve.

[tammet]

For an example of a class of problems WhiteDB is particularly well suited for, please have a look at Enar Reilent's Ph.D thesis "Whiteboard Architecture for the Multi-agent Sensor Systems" pointed to at the front page as http://whitedb.org/doc/Reilent_phd_thesis.pdf. It covers the issues and possible solutions in situations where a number of different software agents communicate information between each other.