Why do so many customers get drawn into the Oracle camp? I do not want to take anything away from the Oracle technology right now – we can have that discussion on another day. Right now, my inaugural blog is about the cost and packaging of Oracle products – primarily Oracle database products.
Are customers getting tired of the Oracle bullying and overpricing? Just yesterday I spoke with a customer discussing the benefits of DB2 v10.5 and Power8 technologies. I shared experiences and successes of other customers discussing how it was common for x86 workloads to require 10X more cores compared to Power servers. This of course directly impacts the TCA / TCO in that it is about 5X less expensive to run Oracle software on Power – you don’t hear this from Oracle sellers and yo wouldn’t expect x86 sellers to say it but they say other things like Power is so expensive – oh, and that it is going away …. so you had better buy x86 and Linux (Oracle Linux of course). This particular customer I was talking with is now running Oracle on x86 now claiming it is killing their budgets. They add a new workload and they have to add more servers which means more Oracle licenses. Upgrade the servers from Sandy Bridge technology to Ivy Bridge and all they do is add more cores but little performance. Of course, this adds more Oracle licenses and cost – Larry has a mortgage to pay….on 19 or so Malibu homes!
I have a large customer who is seeing upwards of 70% compression in their DB2 v10.5 environment – up from 60% when they first migrated from Oracle several years ago. It reduced their licensing and maintenance cost significantly not to mention the cost of storage and storage expansion. They have said it has given their DBA’s a quality of life they never had before. Their enterprise database environment is more stable now than it was before when running Oracle.
What is amazing is when you look at all of the technology that comes with DB2 Advanced Enterprise Server Edition (AESE) which would be in the class of product like Oracle Enterprise Edition but includes all of the features that you pay for with Oracle. Plus, DB2 also includes the first year of maintenance with each year afterward just 20% not 22% like Oracle which you pay in the first year as well as the cost of all the software licenses. Oracle will say that AESE is more expensive but the reality is they charge based on PVU which is 70 for a 2 socket, 100 for a 4 socket (this would be equivalent to the 1.0 for Oracle if normalized) and 120 for the Enterprise class servers (6 & more sockets whether x86, SPARC or Power). This PVU rating system is designed to charge based on the consolidation and feature capability. The best way I can say it is that the 2 socket is Entry Level, the 4 socket is mid-range and the high-end is like the Lexus of servers in that you expect significant RAS, performance, consolidation, virtualization and other features not found on the smaller models.
You calculate the cost by taking the number of cores needed for the workload times the PVU rating times the PVU cost of the product, in this case DB2 AESE v10.5. Oracle is different as they favor platforms that are either their own servers or servers like x86 that use more software licenses – that’s a company you can trust!
DB2 includes their version of RAC, Active Data Guard, Compression, Tools, Tuning and many more items not to mention IBM’s column in-memory product called BLU — all of the items that you pay for with Oracle but are included with DB2 v10.5.
See the complete feature list for AESE at the following link to compare each DB2 v10.5 Edition. From the Advanced Enterprise Server Edition discussed here to the Community Express Edition.
Example List of Features:
- DB2 Database Partitioning Feature
- IBM DB2 pureScale Feature
- Row Compression
- Adaptive Compression
- Intelligent Mining
- Workload Management
- Continuous Data Ingest
- Change Data Capture (CDC)
- Unstructured Text Analysis
- Cubing Services
- Accessing federated data in DB2 for i or DB2 for z data servers
- Accessing federated data in non-IBM data servers, except accessing federated data in Oracle databases through SQL Warehousing Tool
- SQL Replication with DB2 for i or DB2 for z data servers
- SQL Replication with non-IBM data servers
- Column-organized tables
- DB2 Connect
- Memory / sockets / cores / storage – unlimited
- DB2 Governor
- DB2 Advanced Copy Services
- DB2 merge backup
- Oracle compatibility
- Query parallelism
- Replication Tools
- Spatial Extender
- Time Travel Query
- And more …
DB2 has different pricing model options – by core using PVU’s, by the size of the database regardless of the number of cores and even RVU which is to tie the number of licenses required, measured in RVUs, to the utilization of the software or the resources the software manages.
HADR which is DB2’s improved version of Active Data Guard (ADG). It is no cost if the remote site is “cold”. It requires a token 100 PVU licenses at the remote site if “warm” and if the site is “Hot”, it would require full licensing. Oracle charges full licensing for warm and hot.
The DB2 AESE licenses also come with 5 licenses of Cognos, 10 licenses of Infosphere and “Use Limited” DB2 Connect licenses.
When Oracle DBA’s are emotionally ready to set aside their feelings (I love Ford) and ready to test drive a Chevy they will find that it is nearly identical and can begin administering a DB2 environment immediately.
Taking these database benefits that come with DB2 v10.5 and factoring in the cost benefits listed below why would customers A) Consider Oracle database B) Continue using Oracle database?
The following table which I borrowed from the blog of Connor O’Mahony shows all of the tremendous features that are included in the AESE version of v10.5 which also includes the first years maintenance. When maintenance starts in year two it is just 20% vs 22% for Oracle.
|Functionality||DB2 Advanced Ent. Edition||Price|
|Core Database||DB2 Enterprise Server||Included|
|Data Compression||Storage Optimization||Included|
|Advanced Security||Adv. Access Control||Included|
|Data Partitioning||Table Partitioning||Included|
|Administration||Optim Database Admin.||Included|
|Development||Optim Development Studio||Included|
|Performance Tuning||Optim Performance Manager||Included|
|Active/Active Rep.||Q-Replication with DB2||Included|
The following table shows the equivalent products from Oracle. Note how most of them have an associated cost plus their 22% maintenance cost just to buy the license. UPDATE: Aug 02, 2014 (Adding details for Oracles In-Memory feature) The latest feature which adds cost is the Column organized feature called “In-Memory” that is built-in to Oracle Enterprise Edition database. Even though it is built in to Oracle Enterprise Edition like BLU is to DB2, Oracle charges and licenses separately at $23,000 per core or with the first years maintenance of 22% that would make it $28,060 per core – ouch! Looking at my original statement prior to this update which follows is that it is both prophetic but sad as I was spot on. “One thing we know is that it WILL cost money plus 22%! Larry has to buy airplane fuel for his airline in Hawaii and that’s not cheap!”.
|Functionality||Equivalent Oracle Software||Price|
|Core Database||Oracle Enterprise Edition||$57,950|
|Data Compression||Advanced Compression||$14,030|
|Disaster Recovery||Active Data Guard||$14,030|
|Advanced Security||Label Security||$14,030|
|Administration||Oracle Enterprise Manager||No charge|
|Development||Internet Dev Suite||$7,076|
|Performance Tuning||Diagnostics Pack||$6,100|
|Active/Active Rep.||Golden Gate||$21,350|
With the cost for all of these “Add-On’s” that many customers use, especially on x86 servers and for sure on Oracle’s ExaData product, Oracle is now just 3X more expensive than DB2. But, each year the annual maintenance cost will be more; $36,916/core for Oracle vs $13,360/core for DB2. Looking at your server options to increase the efficiency of the software investment there are two options – x86 and Power. I am not a “Z” guy so feel free to tell me it is a viable option as well – I’m happy to learn more on this.
Oracle will claim SPARC but they have a hodge podge of virtualization and a history of changing platforms. T5 is a respectable chipset but unless Oracle can drive significant revenues with the platform I am not sure it has a long term future. I see them mostly using SPARC and Solaris for it’s “mindshare”…it’s fanboi’s – hey, I’m a Solaris fan by the way. I worked at Sun for 10 years and was an instructor in the U.S. Army at Ft Huachuca focusing on SunOS and Solaris in the mid 90’s. Oracle’s focus (today) is on producing inexpensive white box x86 servers as a delivery vehicle for voluminous quantities of expensive software. They recently entered into an agreement with Dell where I expect Dell to begin providing their no-innovation x86 servers to Oracle as part of that partnership. Win-win for both as Oracle can shed any engineering and manufacturing cost while Dell increases their volume business.
Intel’s x86 with Linux delivers improved stability and scalability for an otherwise deficient platform compared to the Power platform. Where Linux makes x86 better, Power makes Linux better for example. Although Linux is an option on Power running in both Big Endian (right ordered) and now Little Endian (left ordered or traditional x86) modes – Oracle does not run with Linux on Power. DB2 does but does not yet deliver the BLU functionality. The traditional commercial operating systems are IBM i (OS/400 – the original integrated system) and AIX. Both of these OSes provide enhanced security, scalability, virtualization and tighter integration with the platform – makes sense since IBM owns the IP for all of it. Unlike x86 where the chipset comes from Intel, motherboard from a vendor, SAN adapter from a vendor, Ethernet card from a vendor, OS from a vendor, virtualization from a vendor, HA from a vendor, so on and so on. Do they work – usually but commodity means their tolerances have to accommodate a multitude of vendors, chipsets and other technology nuances. Probably runs great most of the time but when it does not it may be a fire drill to get 10 vendors together on a call to troubleshoot.
Oracle charges for all cores on x86 times a licensing factor of .5. On Power, you only pay for the cores used for Oracle times a licensing factor of 1.0. Oracle isn’t afraid to manipulate these factors to benefit their sales efforts as seen by them lowering a T series generation to .25 and increasing Intel Itanium from .5 to 1.0. Even while they claim to have the “World’s Fastest Processor” with the T5 which is a joke and a perfect example of the smoke and mirrors perpetrated by the massive Oracle marketing machine they have not increased it to 1.0 for example to match Power if it is so powerful. The licensing factor table is either based on a foundation of trust or mistrust. I’ll let you decide for yourself which it is.
I will spend more time on future blogs discussing the efficiency of Power servers that contribute to licensing differences between x86 that I have seen having a core ratio of 10:1 in but more likely 4:1 is common. Yes, 100 x86 cores to 10 Power cores. It’s based on the workload, utilization and architecture in the end. Depending on the server which is hosting that workload like a enterprise class Power7+ 780 this is not just plausible but demonstrable. With Power8, which was just introduced this past June 2014, it is delivering roughly 2X the performance per core over x86. I say roughly as there are several key benchmarks like TPC, SPEC, Oracle, SAP, Linpack and others. I like the SAP S&D 2 Tier benchmark as it tends to exercise the server with a load often similar to customer workloads.
I will purposely not cite the exact numbers – go look them up. I have looked at them 10,000 times. I just want to give you the general numbers. IBM released a 24 core Power8 result in April 2014 which is a 2 socket server that delivered around 21,000 users and ~115,000 SAPS. Not bad. NEC released a 60 core Intel Ivy Bridge EX E7 v2 result which is a 4 socket server that delivered 20,800 users with 114,700 users. The Power8 has 60% fewer cores but comparable results. There are also other 60 core results from IBM, Dell, Cisco, HP, etc that have more users and SAPS. Not a whole lot more but more. The point is that Power8 is a beast.
If we were to use the performance results above from the SAP benchmark for Oracle there would be 60 x86 cores and 30 Oracle licenses. Take 30 x $204,716 = $$6,141,480. For the Power server, there are 24 Power8 cores and 24 Oracle licenses. Take 24 x $204,716 = $4,913,184 which is $1.2M less for Power8. One point of clarification is on the Power server. If you are not running Oracle RAC the cost would be the cost of the database only at $57,950 per core which lowers both the TCA and TCO even more. Example of 24 licenses: 24 x $57,950 = $1,390,800.
Update: Added a table to make the values above easier to visualize
BUT….But, that isn’t the real story is it? Is it Oracle? Is it x86 sellers? I’m about ready to tell the big secret here to bust this open – this will be bigger than when Geraldo opened Al Capone’s safe with America watching in 1986. The real story is based on the architecture of the solution. Customers do not and would not deploy a single x86 server. They are inherently unreliable lacking serviceability features. They depend on software clustering like Oracle RAC to compensate for their deficiencies. This means more servers, more hardware like adapters, switches, switch ports, software, maintenance, rack space, cooling, advanced skills, etc, etc.
For the above x86 pricing example, that means there would be 2 x 60 core servers totaling 120 x86 cores or 60 Oracle licenses (are you sitting down?). That is 60 x $204,716 = $12,282,960. Oh My Gosh! Seriously!!! 2 little old x86 servers that are supposed to be inexpensive! Good enough technology! The next part of the secret is that Power servers are inherently reliable and serviceable. You typically do not need multiple servers as it is more common to deploy them in standalone configurations for each workload. Yes, businesses will decide to invest and use something like PowerHA which can be around $3,000 per core vs Oracle RAC at $23,000 per core (that is without the 22% annual maintenance) but even if we double the Power8 number that makes it $9,926,368. Note the delta between the x86 and the Power8 Oracle pricing is starting to grow – and this is just for presumably 1 workload – okay, maybe 2 (lots of factors here). Not a stick of hardware has been bought yet.
As Paul Harvey says, the rest of the story is that the Power8 server won’t license all of those cores but the x86 will. That is how it is done. For the Power8 server, they will be sized for the workload – pick a number – let’s say 50% representing a 50% utilization. That is 12 cores are licensed for Oracle out of the 24 active in the server. That is now $2,456,592 per server. Note: Remember if you are using just 1 Power server, with no RAC and there are just 12 Oracle licenses it would be 12 x $57,950 = $695,400. If you used Oracle RAC though, that brings the cost to $4,913,184 for both servers. I don’t suggest using RAC if high availability is needed unless you have an availability requirement around 5 minutes or less. PowerHA can support 5 minutes and less but if 5 minutes is the most your business can tolerate then RAC may be required. If 5 minutes or more is fine then PowerHA is an extremely mature, robust product that is easily deployed with lots of available skills – look for online resources from Shawn Bodily and Michael Herrera.
# of x86 servers
Total # licensed Cores in Solution
1st Year Oracle TCA
Total 1st Year Oracle TCA
The saving grace for x86 is that Power is expensive, right? That is what is written all over the blogs! Some x86 or software seller is out there right now claiming this so it must be true. Or, they could be ignorant of the facts, lying or both – I’m going with both as I have actively seen / heard it. This is unfortunate as I try to have a sense of humor with my work even though I am passionate but as a Christian I also must live my life with balanced scales (Proverbs 11 or 20 – take your pick) which is frustrating to work in a industry built on exaggeration that encourages overstating capabilities just to get a purchase order.
Reality is – since Power7 the price / performance for Power has gone down. Price went down 1/2 while performance went up by 4. Not 4X per core but by form factor. With Power6, what was a 16U server with 16 cores is now with Power7 a 2U or 4U server with 16 cores. That previous 16U server could now provide 64, 96 or even 128 cores. Performance did go up per core as well – depends on what model we are comparing what to what. Not relevant right now. With Power8 we are seeing even more performance in a footprint. That 2U and 4U server are now delivering up to 24 cores for the entry level models while performance per core is also up almost 2X over Power7 with the cost down even more. With the Linux only models on Power8 they actually have x86 price equivalency. The models that run Oracle which implies AIX are “price comparable” to x86. When you see the pricing of Oracle above, the cost of the servers begin to be less significant. What is the price of that 60 core x86 server? I’d estimate the Power8 server is about the same price or maybe 25 – 40% more. Depends if it is a blade product where some of the server cost is shifted out of the blade into the chassis. Power pricing includes the OS and virtualization whereas often the cost of x86 is viewed just at the hardware cost of acquisition. With x86, if you need to add more workloads what will you do? Add 2 more servers and $10M more in Oracle licensing costs + 22% for maintenance every year for the rest of your life – Larry needs a new yacht – every year! With the Power8 server, you could add another 12 core workload on the same server. So, you don’t HAVE to buy another server, you can provision the new workload in minutes AND the software licensing is significantly less.
Back to DB2, if you look at using DB2 instead of Oracle on the Power8 server you will get more performance per core which means you probably won’t need 12 cores but maybe 6, 7 or even 8. But, for purposes of our discussion let’s stay with 12. 12 x $66,800 = $801,600. If you choose to use clustering like RAC then go with DB2’s pureScale which is actually more reliable, scalable and efficient and included. So, the cost for 2 servers is $801,600 x 2 = $1,603,200. (I should add that the DB2 pricing for this 24 core model should actually be less since this pricing is built on a PVU rating of 100 whereas the 2 socket 24 core server has a PVU rating of 70. That is less than the cost of a single server running Oracle. Plus, the annual maintenance is just 20%. Now, you can replicate the database using HADR’s log shipping feature that is included with the DB2 license to something like a IBM Managed Solution Provider. This cost would be $66,800 for 100 PVU’s.
*Reminder the pricing used here relies on a normalized PVU value of 100 to make it comparable to Oracle’s license factor of 1.0 for Power. The 2 socket server used in the table above would actually have a PVU rating of 70 which would lower it’s license cost even more increasing the delta between DB2 and Oracle.
What I hope you get out of my inaugural blog is that you don’t have to continue feeding the Oracle machine unless that is what you want – if so that is fine. If you want options, then consider DB2. Either way, there is only 1 platform to run your enterprise workloads that scale-out or scale-up with maximum efficiency, security, performance, reliability, availability (hold on – not done yet), serviceability, portability, flexibility, value and quality of service – Power servers.