I remember getting the call from Lauren Klein; "Toozs, we want to capture your ravings and rantings for posterity on the internet, in an HDS community forum." "Really?, How do I do that? What do I write about?" and so the painful (for her) process began of extracting stuff out of my head and putting it down on paper.
I really enjoy working on messaging around the value of Hitachi, both for now and for the future. It has been a really satisfying to have customers and potential customers use my material to help them understand the value that our engineering and our brand brings to their business.
My work with Thomas Stutesman's team (Robert Reichl, Werner Still, Pete Gerr et al), on the UCP 4 Oracle has had outstanding results. One of the customers in question, a South African Telco client is now a fully referencable customer, who takes pride in the dramatic savings he now gets from his UCP system. I have paraphrased a reference call he had with another customer below.
For me the Community has surpassed what we had hoped it would in year one. We are getting many messages and questions on "how to", on "what next" and on details of "who can I talk to?". I have personally had over 400 hits on my stuff, which is extraordinary, who would have known at the outset that so much of this stuff would be of so much use to so many people.
So, in the spirit of keeping it interesting and relevant, here is the promised stynopsis of the meeting between an existing customer who has an inkling that UP for Oracle is a great thing, and one who already knows:
A South African Telco client networking team ran several tests that were provided by the "pro-Exadata" people. There were already 2 full and one ¼ Exadata racks from Oracle in the Data Center. HDS supplied a single ¼ rack. In running the tests from the network monitoring systems (billing and others) they needed to reduce the time for reports from >2 days to less than an hour.
Full rack Exadata = 7 minutes
¼ rack Exadata = 72 minutes
¼ rack UCP = 20 minutes
When comparing the cost of the ¼ racks, the South African Telco client said that the price of Exadata ¼ rack was same as UCP ¼ but the 5 year TCO (price + licenses + maintenance) was double the cost. Licensing is based on the number of HDD and cores. UCP does not need as many disks therefore has a MUCH lower overall cost. This was a straight cost for cost comparison, inclusive of ELA; the South African Telco client Networking took their costing from the Procurement Company (VPC) team who included all the ELA and discounts offered by Oracle to the client. When comparing the Exadata full rack with the ¼ rack UCP There is a 13 minute difference in time taken to do the calculations. This is significant when comparing with 20 minutes, but it represents an 80% higher TCO over 5 years than the UCP. This is not a justifiable cost when comparing with the > 2 days for the original SQL Query; the difference (> 2 days to 20 minutes) means it is good enough!
When comparing the 2 x ¼ racks the South African Telco client found that the 3-fold increase in performance was because the data was very unstructured and in-memory, maximizing the use of FusionIO cards. These results were consistent across all the SQL queries in spite of what they were, so UCP ¼ was always 3 x faster than the equivalent Exadata. Ergo 3 x speed at half the price; so a massive price performance benefit.
The South African Telco client Networking can no longer justify the continued use of Exadata.
Other salient points:
· The UCP Scalability was really impressive, much better than Exadata.
o All the upgrades and changes performed on the system during the PoC were 100% non-disruptive upgrades (adding blades and components)
o UCP was consistently 3 times as fast as Exadata ¼ rack yet with far fewer (licensable) cores
o UCP uses native Oracle tools to manage the DB instances.
o System extracted 70TB datamart actually processes 25 - 30TB
o There is a perceived benefit and differentiator from the use of LPAR, but it was not used in these tests because the system was so large didn't need it.
· Significant cost savings:
o The South African Telco client could add license costs as required and as the system changed, not pay upfront
o Far fewer HW licenses.
o Far lower TCO throughout the range (S, M and L).
· Hitachi Services were better and less expensive than Oracle on an on-going support basis:
o Oracle has a lack of skills and required far more people to set up than HDS, and they appeared to struggle, taking more time
o HDS set it all up for us and it was very smooth. They participated in the whole process.
o Exadata can do nothing without engaging Oracle Professional Services (costly!)
o HDS were always at the end of the phone and on-site. Oracle support was not so great, and our South African Telco client still has to pay support fees for this service which is a big consideration for our client and another of the causes for moving away from Oracle.
· Versatility from UCP :
o Oracle is an engineered system, cannot tune it whereas UCP you can!
o The South African Telco client can also run Sybase (or other DB systems) on UCP not just Oracle, Exadata only runs Oracle.
o Have bigger plans for the system to source more data and place it in there for this application
o Also include other projects over the coming months, as it is now a "Standard"