Nick Toozs-Hobson

UCP4 Oracle; what makes us different?

Blog Post created by Nick Toozs-Hobson on Oct 1, 2013

You may know that Hitachi Data Systems has invested huge amounts of time and resources in develping a converged platform for Application serving called Unified Compute Platform. It appears very similar to some of our competitors offerings for in-memory data handling or for running large virtual server farms or for handling varying and various workloads in between. The UCP though has one or two tell-tale differences that enable it to run all these workloads, at the same time and in the same rack on a consolidated platform. It is also where the similarities between us end.

For now, though, I'd like to focus on why (and how) we have probably the best value for money. Oracle have seen significant traction for their thoroughbred and rightly so, but before commiting to that route, take a look at UCP4 Oracle, compare, and we think you might just change your mind!

We have recently run some customer proofs of concept in EMEA for two very well-known Global brands. In both cases the proof was extremely evident. In both cases Oracle supplied an Exadata Full Rack/8, and when the tests were run, although there was a performance difference, it was so small and the difference in price so high, that the customer had no hesitation in switching to UCP4 Oracle. The customers also had significantly lower post-installation running costs .

 

I have attached data that shows both the performance and the architectural differences. This shows that the reduction of time to complete SQL queries from the current system to the proof of Concept systems is astounding. It turns days and hours into minutes and seconds. UCP4 Oracle achieves good enough performance by comparison but in a much smaller sized configuration. How come? UCP uses some of the underlying technologies of both the Hitachi Servers and Storage components (detailed below) which when combined allow the architecture to deliver the performance requried. We have highlighted the differences in HW to illustrate where the license savings occur because Oracle charges a license fee for the HW components on which it runs. By being able to reduce the components required to perform the same tasks in similar time frames we do so deploying fewer licensable parts.

So, how do we do this and what makes us different? In our x86 servers we have applied what are Mainframe-like Services that no other x86 architecture can do (today) and added Hitachi Flash to our existing controllers that work like an All Flash Array (simliar to a Violin or Extremio, except inside the same domain).

  • Symmetric Multiple Processing (SMP) to combine workloads and cluster CPU. This reduces the number of cores used and these are counted per CPU ergo, the less CPU's there are, the fewer the cores, the fewer base licenses (in both the attached instances we needed 80 and Oracle needed 160)
  • Hybrid IO to split, push and manage the workload in different parts of the infrastructure. In high read environments we send reads to NAND Flash cards connected to the CPU via a PCI slot. (and in the Storage Controller and array) we can deploy Hitachi Accelerated Flash (HAF) technology to increase the velocity of writes. By deploying these unique technologies we end up using far fewer HDD to perform the same workload. Oracle has a license "per HDD deployed", again, the fewer there are the less we pay
  • Logical Partitioning (LPAR) or the ability to isolate workloads on a server. LPAR enables us to build a cold failover capability which, when combined with Oracle Data Guard for High Availability / Business Continuity, removes the need to use Real Application Clusters (RAC)  which are very expensive and CPU intensive licenses.

The next obvious upshot of fewer "moving" parts, is that there is less to power up and cool down too, and as UCP has a smaller footprint than Oracle, the facitlities manager can smile as well as the procurement team!

We looked at the value of the "promise" that UCP is easier to deploy and investigated how long it actually took to set up and start working. You will see that there is a table showing deployment times for these tests. On day one Oracle can plug in the HW and get it up and running faster than UCP, because Oracle pre-loads certain SW in their Distribution Centre. When we know the customer and they have an "Image" of their desired environment in production, we do this in our Distribution Centres too, and this will save a day or two on site. In these particular POC's we could not do that because we needed to be flexible, these UCPs were for multiple tests.

 

Having plugged it in and got it going we had to install the OS and then load in the applications and the data, well that's basically the same for everyone, or is it? Well, actually, no it isn't; we found that there are two key criteria for the data sets. a) the size of the data, the bigger it is the longer it takes, that just because it takes what it takes to load it all up, and b) it can take up to 4 manweeks to "tune" the data set for the Exadata just to get the data in a position to run the SQL queries for the reports required. At both of our customers, Oracle Professional Services were called in to make it work. This adds to the time and cost for a customer to deploy.

 

(As an aside, we ran a test earlier this year with a European Global Systems Integrator, where we had UCP Pro in and working in 3 days and about 9 mandays of effort we had all the apps installed and running. Admittedly, this wasn't an Exadata comparison, but we beat VCE by 50 mandays or more of effort!)

Finally, and possibly the biggest single architectural difference, is that we are able to consolidate many versions of Oracle (from 9i through 12c) on the UCP4 Oracle by deploying a blend of the unique technologies mentioned above. And we have investigated the need for running SAS to do data extrapolation to populate the Data warehouse in the same rack. We have spent time at SAS discussing how to do this. We showed them UCP4 Oracle and how it works and they have agreed two things a) it will run in the same rack (on separate LPAR or blade) and b) they are in the process of  "certification" to corroborate this. We believe that this is unique among x86 server vendors.

 

I really hope you this is useful material, and maybe change any thoughts you have about Hitachi Data Systems ability to serve applications running in Oracle.

Attachments

Outcomes