Simon Haslam's Oracle Fusion Middleware blog

What you need to know about the new ODA X5-2

Today, as part of the "Next Generation of Oracle Engineered Systems" webcast,  Larry Ellison launched the new X5 systems. This bullishly-titled post attempts to summarise what's new specifically with the ODA X5-2, and what's most important, especially for those using ODA Virtualized Platform (ODA VP) to build entire Oracle infrastructures as an appliance.

We've known since last September when Intel released the Haswell-EP processors (the E5-2600 v3 models) that there would likely be refreshes to many of Oracle's engineered systems. However for this year's ODA refresh there have been far more changes than the previous one (which was just the processor update and fibre option). 

Oracle Database Appliance X5-2 

Summary of Changes 

Here are the most significant changes in the ODA X5-2, as compared to the previous X4-2 generation, biased towards my perspective of running Fusion Middleware products, and associated databases, on ODA VP for O-box:

  • Extra SSD on shared storage ("ODA Flash Accelerator") to hold some database data ("ODA Flash Cache") and ACFS metadata ("ODA Flash Files")
  • 40Gb/s InfiniBand for interconnect between server nodes
  • DDR4 memory with the option to upgrade to 768GB per node, so 1.5TB total
  • SAS3, which runs at 12 Gb/s - I assume/hope this is for connections to server disks, internally within the array(s) and between the arrays and servers

 Less important changes:

  • Slight price increase from $60k to $68k (additional storage shelf remains at $40k)
  • PCIe RAID controller cards to perform the server system (and ODA Base) disk mirroring, rather than using software RAID

In this article I'm primarily talking about hardware changes (although ODA Flash Cache etc must have a software component).

The details

Now let's look more closely...

ODA Flash Accelerator

There have been a few names floating round in this area, but essentially the ODA now has 3 categories of persistent storage instead of 2: SSD for REDO, SSD for frequently used data, and HDD for bulk data, backup/recovery and VM storage. My understanding  (currently vague!) is that "ODA {Fast} Flash Cache" provides read caching of data blocks, somewhat like the database's main memory buffer cache, but because it is available identically to both RAC nodes, blocks don't have to be transferred over the interconnect (sounds more like OPS!) - I don't really understand how LMS must work, but Oracle seems very excited about it. Note when listening to the webcast about the new Exadata X5-2 flash storage server it sounds like the PCIe flash (now as NVMe SSD) has moved from Exadata compute nodes to shared storage so maybe the ODA approach is similar (but don't quote me on that!).

It seems you can also just put some or all of your data blocks in the +FLASH SDD disk group, and the ACFS metadata ("ODA Flash Files") is stored there too to give "near-native filesystem performance". This latter feature shows how seriously Oracle is taking ACFS storage of databases (default for ODA 12c) and should have benefits for VM users too.

So these new 4 x 400GB SSD (800GB, presumably mirrored), with very respectable IOPS from what I've heard mentioned, look like a useful performance improvement for portions of middleware repositories, for example for some of the SOAINFRA schema on an O-box SOA system running, say, a high proportion of orchestration of async services and human workflow. This is before we consider network connectivity between the middle tier and the database...

Networking

In many ways the most surprising news about the ODA X5-2 is that the ODA's interconnect, rather than being 10Gb Ethernet, is now InfiniBand running at 40 Gbps - a key differentiator for the Exa* machines. It has the same cross-over/switchless topology as previously as there are only two connected server nodes.

I don't know that much about InfiniBand but can believe that it gives us a more modern, much lower latency interconnect than ethernet. I am assuming that the ODA provisioning will sort out all of the InfiniBand configuration, and later patching, for me. Juan Loaiza on the webcast was talking about InfiniBand optimisations for running OVM on Exadata, so let's see what Oracle's InfiniBand implementation ODA VP is like.

Finally, if you want the optional fibre connectivity, introduced with ODA X4-2, you can do that but only by sacrificing the InfiniBand cards and switching the interconnect back to copper. I think I'd prefer to find an alternative. Perhaps we will see some on-board SFP+ ports in future (perfectly feasible, and some manufacturers have offered them though maybe not recently) which would resolve this issue far better. 

Memory Upgrades

Whilst previously there was an unpublicised, but little known, option to upgrade ODAs from 2 x 256GB to 2 x 512GB this was relatively expensive to do as the original 16 DIMMs had to be thrown away. Instead the new X5-2 servers have 24 DIMM sockets and the base model has only 8 populated, but with the largest modules currently available (32GB). Therefore upgrading to 512GB per node is simply adding 8 DIMMs to each - each 256GB upgrade kit is $11k (list), so a 2 x 512GB ODA X5-2 will cost you $90k list, or 32% more than the base model. For ODA VP implementations, like any of the O-box SOA medium and larger configurations this, is likely to be a very worthwhile upgrade.

Miscellany

There are two new features I haven't yet mentioned and which I suspect are aimed at database-centric consolidation projects:

  • New Intel E5-2699v3 18-core processors giving have a total of 72 cores!
  • New storage array/shelf using 3.5" disks that increase the raw HDD capacity by 3.5 times (from 21TB to 32TB double mirrored). Eagle-eyed ODA fans will have already noticed from the photo that this array is 2U deeper than previously, giving an overall ODA depth of 6U (and 10U if you choose 2 arrays).

Readers might be surprised that I'm not rating these changes more significantly - this is primarily because they don't address mid-market pain points in my experience. For example you might think that a machine running lots of VMs would need many cores but 48 modern Xeon hyper-threaded cores are a lot! One concern I did have was that such a dense processor may perform workloads on fewer cores relatively poorly compared to, say, a 12 or 8 core processor but, reading the benchmarks and discussions online, it appears these fears may be unfounded (a topic I'll come back to another day).

Also I've primarily focussed on hardware in this post. ODA fully supports In Memory and Multi-tenant database options (InfiniBand is needed for clustered IMDB) which will be relevant topics to some people. Also the Enterprise Manager 12c ODA plug-in, as I've mentioned earlier on the O-box blog, is on its way and Oracle are now talking about a new wizard to automate provisioning of a full EM 12c system (OMS and OMR), rather than using the template and white paper.

Closing Thoughts

In summary there are several new features in the ODA X5-2 that I like - more memory and ODA flash caching in particular. The extra (more than triple) storage capacity will suit some people, e.g. running an O-box WebCenter Content appliance on ODA, plus of course those looking for database consolidation.

It's also very interesting when you start looking at how ODA compares to the smaller (now incrementally upgradable) Exadata models. Whilst HCC and Smart Scan are restricted to Exadata these are mainly BI/DWH features (there may be others I'm missing!) and perhaps many people don't really need them (especially with options like IMDB now available). With its increased performance and capacity, a relatively quick investment write-off, super-fast deployment time, and not "needing a dedicated (hu)man to run your system", I think ODA could be very attractive proposition to many more customers now.

...not that you probably expected a different verdict from an ODA-fan like myself!

For  further information, and hopefully an ODA X5-2 data sheet soon, see https://www.oracle.com/databaseappliance

 

Footnote 22/1/15

If you liked this article and will be in London on 26 February please come along to the free Club Oracle event, hosted by Red Stack Tech (fka e-DBA) and Jonathan Lewis, where I will be discussing ODA X5-2, and Appliance Manager 12.1.2.2, in more depth:

The Oracle Database Appliance (ODA) - an integrated platform based on two Exa-style compute nodes and direct attached storage - has been evolving rapidly during its relatively short life. In this session I will focus on some of the interesting technical aspects of the ODA (such as sizing, huge pages, datafiles on ACFS, virtualisation and management) and anything new that might or might not have appeared in January or February.

Registration: http://cluboracle-feb.instapage.com/

Comments:

Post a Comment:
Comments are closed for this entry.

Search this blog

About me
Oracle ACE Director (Middleware and SOA)
Presentation downloads



I'm speaking at Oracle OpenWorld


UKOUG Ambassador Partner
Oracle WebLogic Server 12c Certified Specialist
Oracle WebLogic Server 12c Certified Specialist

Links

Feeds