Comments on: HBase vs. Cassandra: NoSQL Battle!

By: NoSQL Is Not SQL And That’s A Problem | CloudAve

NoSQL Is Not SQL And That’s A Problem | CloudAve — Sat, 06 Mar 2010 02:37:36 +0000

[...] data persistence and access solutions that challenges the long lasting legacy of RDBMS. Competition between HBase and Cassandra is heating up. Amazon now supports a variety of consistency models on EC2.However none of the NoSQL [...]

By: HBase vs. Cassandra: NoSQL Battle! | Road to Failure « The other side of the firewall

Wed, 24 Feb 2010 23:04:08 +0000

[...] February 24, 2010 at 6:03 pm · Filed under Cloud computing ·Tagged Databases [From HBase vs. Cassandra: NoSQL Battle! | Road to Failure] [...]

By: Bradford

Bradford — Thu, 24 Dec 2009 05:21:32 +0000

Thanks for the feedback. I’m going to do another end-to-end test with the next release of both products~

By: Jeff Darcy

Jeff Darcy — Tue, 24 Nov 2009 13:52:17 +0000

Comparing an A+P data store to a C+P one is a tricky exercise at best, but some of the claims here are particularly disingenuous. For example, the region-server and NameNode SPOFs are major issues for a lot of people, but they’re glossed over quickly while the inaccurate “depends on very fast fiber” gets repeated four times. In fact, depending on how you allocate keys and how you set W/R, you’ll usually be able to get an immediate answer locally despite failures. Even when there are too many failures for that to happen, getting a slow and/or stale answer from China is still (for many applications) preferable to getting no answer at all because of a SPOF in a system that as of today has barely any tolerance for partitions at all.

Does that seem a bit unfair? Of course it would be, in a vacuum. C+P and A+P are both valid tradeoffs, suitable for different situations. If Cassandra is going to be repeatedly slammed for compromising on C, though, then it seems quite fair to point out that HBase compromises just as much on P. I’ve been among those pointing out the holes in some of Cassandra advocates’ claims, but that doesn’t mean I have much patience for equally invalid claims from critics or detractors.

BTW, your definitions of A and P are a bit off from those used in the CAP theorem (see Gilbert and Lynch’s proof paper for the most rigorous definitions), and “vector clocks mean latest timestamp wins” is extremely misleading at best.

By: schubert zhang

schubert zhang — Fri, 06 Nov 2009 16:27:50 +0000

To compare two different architectures, I think it should be objectively and theoretically. Some temporal un-ready features should not be the subjects. Such as the “multi-table” feature, it seems not difficult to be implemented.

It is very interesting that many people doubt Dynamo. Even guys from Facebook: http://jsensarma.com/blog/2009/11/dynamo-a-flawed-architecture-part-i/

I think each architecture is designed for its usage.

By: Twitted by abrdev

Twitted by abrdev — Fri, 30 Oct 2009 20:53:25 +0000

[...] This post was Twitted by abrdev [...]

By: Twitted by pogrebnyak

Twitted by pogrebnyak — Fri, 30 Oct 2009 11:09:27 +0000

[...] This post was Twitted by pogrebnyak [...]

By: Twitted by tkang1

Twitted by tkang1 — Fri, 30 Oct 2009 07:57:26 +0000

[...] This post was Twitted by tkang1 [...]

By: Twitted by ungerik

Twitted by ungerik — Fri, 30 Oct 2009 05:01:37 +0000

[...] This post was Twitted by ungerik [...]

By: Twitted by andreisavu

Twitted by andreisavu — Thu, 29 Oct 2009 23:51:35 +0000

[...] This post was Twitted by andreisavu [...]