Exadata: What’s Coming

This is based on the presentation Juan Loaiza gave regarding What’s new with Exadata. While a large part of the presentation focussed on what was already available, there are quite a few interesting new features that are coming down the road.

First of was a brief mention of the hardware. I’m less excited about this. The X4 has plenty of the hardware that you could want: CPU, memory and flash. You’d expect some or all of them to be bumped in the next generation.

New Hardware

This was skated over fairly quickly, but I expect an Exadata X5 in a few months. The X4 was released back in December 2013, first X4 I saw was January 2014. I wouldn’t be surprised if Oracle release the X5 on or around the anniversary of that release.

Very little was said about the new hardware that would be in the X5 except that the development cycle has followed what intel has released, and that cpu cores have gone up and flash capacity has gone up. No word was said on what CPU is going to be used on the X5.

The compute nodes on an X4-2 have Intel E5-2697 v2 chips this is a 12 core chip running at 2.7GHz. I’d expect an increase in core count. The X3 to X4 transition increased core count by 50%. If that happens again, we get to 18 cores. There is an Intel E5-2699 v3 with 18 cores but that’s clocked at 2.3GHz.

However, I think I’d be less surprised if they went with E5-2697 v3 which is 14 core chip clocked at 2.6GHz. That would be a far more modest increase in the number of cores. The memory speed available with this chip does go up though – it’s DDR4. Might help with In Memory option. I also wonder if they’ll bump the amount of memory supported – this chip (like the predecessor) can go to 768GB.

As I said, it was not mentioned which chip was going to be used, only that Intel had released new chips and that Oracle would be qualifying their use for Exadata over the coming months.

New Exadata Software

There was a bunch of interesting sounding new features coming down the road. Some of the ones that in particular caught my eye were:

The marketing friendly term “Exafusion”. Exafusion seems to be about speeding up OLTP, labelled as “Hardware Optimized OLTP Messaging” it’s a reimplementation of cache fusion. Messages bypass network stack leading to a performance improvement.

Columnar Flash Cache – This is Exadata automatically reformatting HCC data when written to flash as a pure column store for analytic workloads. Dual formats are stored.

Database snapshots on Exadata. This seems designed with pluggable databases in mind for producing fast clones for dev/test environments. Clearly something that was a gap with ASM as used on exadata, but ACFS does snapshots.

Currently the latest Linux release available on Exadata is 5.10. Upgrading across major releases is not supported – would have required reimaging. Not a pretty prospect. Thankfully Oracle are going to allow and enable upgrading in place to 6.5.

Some talk about reducing I/O outliers both in reading from hdd and in writing to flash.

Currently with IORM you can only enable or disable access to flash for a particular database. Full IORM seems to be coming for flash.

Final new feature that caught my eye was the long rumoured Virtualisation coming to Exadata. OVM is coming. The ODA for example has had VM capability for some time, so it’s in some ways an obvious extension. I’m expecting with the increasing number of cores lots of smaller organisations may not actually need all those cores and might think even if they could turn unused ones off, it’s a waste buying that hardware and not being able to use it.

I’m hoping to NOT see OVM on an Exadata in the wild anytime soon.

Software on Silicon

One final point almost tucked out of site, was that Juan had a little bullet point about “software on silicon”. Now this has me confused. My understanding is that when Larry was talking about this, it was specifically SPARC. That I can understand as Oracle controls what goes on the chip.

Ignoring the SPARC Supercluster, there is no SPARC on Exadata. So that leaves a closer collaboration with Intel or moving to SPARC. Collaborating closer with Intel is a possibility and Oracle had first dibs on the E7-8895 v2 for the X4-8.

I can’t imagine changing the compute nodes to SPARC that wouldn’t make sense. But “software on silicon” is a bit like offloading…

Exadata software is definitely keeping moving forward and the difference between running Oracle on Exadata compared with non-exadata is growing ever wider with each “exadata only” feature.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s