Oracle recently announced its latest iteration of Exadata – X5-2. It includes a refresh of the hardware to the most recent Xeon® E5-2699 v3 CPUs. These new CPUs boost the total cores count in a full rack to 288. This is higher than the current 8 socket “big machine” version X4-8, which has only 240 cores.
But the most exciting part is the all flash version of Exadata. In the previous generation – X4 – Oracle had to switch from 15K drives to 10K drives in order to boost capacity from 600 GB to 1200 GB per hard drive to keep disk space higher than flash cache size. At that time of X4 announcements, we were already wondering why Oracle was still offering high-speed disks and not switching to all flash, and now we know why. Because that type of high-performance flash wasn’t quite ready.
Maintaining high IO rates over long periods of times needed some changes to the ILOM in order to maintain cooling fans speed based on many individual temperature sensors inside the flash cards (details). Removing the SAS controller and using the new NVMe connectivity resulted in much higher bandwidth per hard drive – 3.2 GBytes/sec vs. the old 1.2 GBytes/sec SAS.
With temperature and bandwidth sorted out, we now have a super-high performance option (EF – Extreme Flash) for Exadata which delivers the stunning 263 GB/sec uncompressed scan speed in a full rack. The difference in performance between the High Capacity and High Performance EF flash option is now much higher. The high-performance option in Exadata X5 is now viable. In Exadata X4 it made so little difference, that it was pointless.
The one thing I wonder with the X5 announcement is why the X5-2 storage server still uses the very old and quite outdated 8 core CPUs. I’ve seen many cases where a Smart Scan on an HCC table is CPU bound on the storage server even when reading from spinning disk. I am going to guess that there’s some old CPU inventory to cleanup. But that may not end up being such a problem (see “all columnar” flash cache feature).
But above all, the most important change was the incremental licensing option. With 36 cores per server, even the 1/8th rack configuration was in the multi-million dollars in licenses, and in many cases was too much for the problem in hand.
The new smallest configuration is:
- 1/8th rack, with 2 compute nodes
- 8 cores enabled per compute node (16 total)
- 256 GB RAM per node (upgradable to 768 GB per node)
- 3 storage servers with only half the cores, disks and flash enabled
Then you can license additional cores as you need them, 2 cores at a time. Similar to how ODA licensing option worked. You cannot reduce licensed cores.
The licensing rules changes go even further. Now you can mix & match compute and storage servers to create even more extreme options. Some non-standard examples:
- Extreme Memory – more compute nodes with max RAM, reduced licensed cores
- Extreme Storage – replace compute node with storage nodes, reduced licensed cores
In conclusion, Oracle Exadata X5 configuration options and the changes it brings to licensing allows an architect to craft a system that will meet any need and allow for easy, small step increments in the future, potentially without any hardware changes.
There are many more exciting changes in Oracle 12c, Exadata X5 and the new storage server software which I may cover in the future as I explore them in detail.