Revisiting 1M Writes per second

My other favorite &quot;scalability&quot; study is from WhatsApp:<p><a href="http://www.youtube.com/watch?v=c12cYAUTXXs" rel="nofollow">http:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=c12cYAUTXXs</a><p>That&#x27;s &#x27;Billion&#x27; with a &#x27;B&#x27;: Scaling to the Next Level at WhatsApp<p>(that walk title was create before the acquisition and was mean to imply message count, after the acquisition it got a secondary meaning).<p>The one thing that is fascinating about it, is how small their team was compared to the volume and complexity of the operation.

I love using Cassandra; it&#x27;s been a dream for analytics. Thank you Netflix et al. for not only moving the project forward, but providing research and commentary like this and others available at <a href="http://wiki.apache.org/cassandra/ArticlesAndPresentations" rel="nofollow">http:&#x2F;&#x2F;wiki.apache.org&#x2F;cassandra&#x2F;ArticlesAndPresentations</a>

Would be great to know the exact data set size (not size in the database). I can&#x27;t get an order of magnitude sense of what I am looking at without that.<p>I know I can divine it from the parameters to stress, but I have no idea if the row keys generated by different clients overlap and I don&#x27;t know the default number of columns nor their size.<p>I think it&#x27;s also important in this kind of benchmark to describe the distribution of access especially for a read intensive benchmark. Without that you really don&#x27;t know what your are looking at. I am a fan of scrambled Zipfian.

So the test was run on a 285 node cluster with 60 clients. It would be nice to know how they arrived at those numbers. Was there some sort of formula used to calculate how large each group should be? How much trial and error was involved.

Couldn&#x27;t help but notice: $398.70 per hour = $9568.80 per day = ~$3.5m per annum. They obviously get a discount...but still. What kind of discount do guys like this get?

This is running Cassandra 1.2.x which is over 18 months old.<p>Here&#x27;s some performance notes on the latest (2.1rc4): <a href="http://www.datastax.com/dev/blog/cassandra-2-1-now-over-50-faster" rel="nofollow">http:&#x2F;&#x2F;www.datastax.com&#x2F;dev&#x2F;blog&#x2F;cassandra-2-1-now-over-50-f...</a>

I wonder where the term sidecar originated from and what is the precise definition? Is this something invented by the netflix OSS people or does it predate that?

How many companies can possibly afford the cost and management pain of running a 285 node database? Few would have 285 servers of any type? So if this is what it takes to get 1M writes on Cassandra that is some poor ROI. 10K writes&#x2F;sec is not impressive.