I'm late to the party on this, but unless I'm misreading things... the author used Riak the wrong way and it failed, and then he used it the right way, and it succeeded.
Granted, as the author says, Basho tries to delay the necessity of "breaking the news" about the flaws of LWW to prospective Riak users, but the author references the Dynamo paper. The Dynamo paper is chock full of information about about siblings and the key role they play. If you read the Dynamo paper, you would know immediately that using a system like Riak without thinking through sibling resolution is akin to lighting your data on fire. This complexity of siblings is the tradeoff you pay for high availability (no silver bullet, etc).
And if your scaling/availability problems are serious enough to even begin saying "we a distributed database like Riak!", hopefully it's not asking too much to have you read the Dynamo paper to understand how these sorts of databases actually work.
( note - we admin a largish riak cluster here at bu.mp )
It makes sense to me to test the default configuration that Riak ships, and even with the caveat of "you probably shouldn't use this", it's failures are an instructive lesson (which is the underlying goal of the series of blog posts, not just benchmarking various databases).
Granted, as the author says, Basho tries to delay the necessity of "breaking the news" about the flaws of LWW to prospective Riak users, but the author references the Dynamo paper. The Dynamo paper is chock full of information about about siblings and the key role they play. If you read the Dynamo paper, you would know immediately that using a system like Riak without thinking through sibling resolution is akin to lighting your data on fire. This complexity of siblings is the tradeoff you pay for high availability (no silver bullet, etc).
And if your scaling/availability problems are serious enough to even begin saying "we a distributed database like Riak!", hopefully it's not asking too much to have you read the Dynamo paper to understand how these sorts of databases actually work.
( note - we admin a largish riak cluster here at bu.mp )