Hi Folks,
It was great meeting so many of you at GothamGo a few weeks ago.
Based on some comments and measurements we've made, we increased the performance of our Go client just a little more.
The particular change was lowering the number of bytes in a NewBuffer. Frankly, I found this surprising, because I think of Go having more per-object overhead, and less per-byte overhead. Here's the (small and simple) changelist :
https://github.com/aerospike/aerospike-client-go/commit/dd95d39bffba3fec9e08b2c7e9a4a728f2612b40I think we still have a few more changes to go, like the change to LIFO object caching instead of FIFO that Channels that David suggested.
We made a nice, friendly version of our client documentation on our website, although the Github repo doc is still there
http://www.aerospike.com/docs/client/go/And sorry for the following slightly sale-y statement, but we're very excited about our performance on Google Compute, and we put out a blog post on Google's blog about our numbers in comparisons to Cassandra. As an open source database used by lots of adtech companies, we think when it comes to clusters, "size isn't everything"
http://googlecloudplatform.blogspot.com/2014/12/aerospike-hits-one-million-writes-Per-Second-with-just-50-Nodes-on-Google-Compute-Engine.htmlAnd best practices for using Aerospike in GCE
http://www.aerospike.com/docs/deploy_guides/gce/and we gotcha covered in EC2 as well
http://www.aerospike.com/docs/deploy_guides/aws/Share and enjoy!