@everyone - thanks for all the comments. i have incorporated the multiple comments about the data loss scenario being incorrect when using vector clocks. i have been thinking too much about Cassandra lately - and it doesn't use vector clocks. that said - i continue to believe that returning stale reads is bad and best avoided and that unbounded stale-ness is not acceptable for many applications. To the extent that this is an avoidable scenario in a tightly coupled environment within a single data center - i consider it to be a significant drawback.
@jbellis - i hope the responses to your comment have convinced you about the problems with the dynamo quorum scheme/read-write protocols. i can convey that the problems i described definitely do exist in Cassandra. Jun Rao has made this point in the Cassandra public mailing lists a long time back and i have a pointer to the JIRA that he filed in my post as well.
regarding Dynamo being an interesting design space and for the academic community etc. That may very well have been the case - but the reality of the situation is that the world is now overflowing the Dynamo clones with people considering them for all kinds of usages. Hey - if it was good for Amazon (and Facebook and LinkedIn) - it's probably good for me! The people trying to use Dynamo clones do not understand all the small details. They don't understand what applications would be safe to write on it and which would not. I hope my posts (imperfect and opinionated undoubtedly) would provide a counterpoint to this sentiment and make users think harder and deeper before they make the leap.
i was also hoping to trigger a discussion (looks like i succeeded). i hope that it takes us to a better design space than what exists currently.
@jbellis - i hope the responses to your comment have convinced you about the problems with the dynamo quorum scheme/read-write protocols. i can convey that the problems i described definitely do exist in Cassandra. Jun Rao has made this point in the Cassandra public mailing lists a long time back and i have a pointer to the JIRA that he filed in my post as well.
regarding Dynamo being an interesting design space and for the academic community etc. That may very well have been the case - but the reality of the situation is that the world is now overflowing the Dynamo clones with people considering them for all kinds of usages. Hey - if it was good for Amazon (and Facebook and LinkedIn) - it's probably good for me! The people trying to use Dynamo clones do not understand all the small details. They don't understand what applications would be safe to write on it and which would not. I hope my posts (imperfect and opinionated undoubtedly) would provide a counterpoint to this sentiment and make users think harder and deeper before they make the leap.
i was also hoping to trigger a discussion (looks like i succeeded). i hope that it takes us to a better design space than what exists currently.