In the seven months since we released Event Hubs we have had a tremendous amount of traction in the market place. We’ve learned a lot running a service at this scale: about our service, about cloud services, and about our customers.
One thing that we’ve found is that most of our customers do not require the scale that we defaulted Event Hubs to. Although we chose a default of one Throughput Unit (TU = 1000 events / 1 MBs of ingress) through customer surveys we chose the default partition count through a less scientific means. The default we chose was 16 partitions for an Event Hubs and the minimum was 8. While customers have been very happy with the 1 TU they have been less happy with the partition counts.
Most customers don’t actually need 16 partitions (which is designed for more than 16 MBps of ingress or more than 1.3 billion events per day). In fact many have voiced that having to read from so many partitions is actually an inconvenience – particularly in their early stages with Event Hubs.
So we have decided to change these defaults. Effective today the minimum partition count for an Event Hub will be lowered to two (2) partitions. The default will be lowered to four (4) partitions.
With this change it will be easier for customers to work with Event Hubs when they have lower scale needs.
Please be aware that if you’re code used a default in the past and then later hard coded a dependency for 16 you could be impacted by this change when you create Event Hubs in the future.
Next week we will also blog about how a single partition can actually handle more than one Throughput Unit of volume. Stay tuned.