Why does Oracle design and building their own Intel based servers ?
X7-2/L is primarily used for Oracle internal cloud demand. But volume ( less than 20k ~ 30k per year ) still does not justify designing and building your own server.
X7-2c ( internal use only) used for OPC has volume forecast of less than 10k to 12k per year.
And volume forecast for X7-8 is even lower ( less than 1k to 2k per year )
Why invest in R & D, manufacturing and supply chain management for such low volume products ?
Why does Facebook design their own servers (granted, they then open source them). Why does Google?
EXACTLY! Any more questions?
Why does Facebook design their own servers (granted, they then open source them). Why does Google?
Why invest in R & D, manufacturing and supply chain management for such low volume products ?
If you worked in the X86 you would be able to see all the BS rolling down from the cloud overlords for such a "low volume" product. And if you worked in HW in general, you would recognize the value of internal engineering versus farming it out and having to deal with the intrigues of external "vendors and partners".
And the answer is that Oracle doesn't have to do the manufacturing and SCM parts. Besides, these days it doesn't take a lot of R&D to design an x86 box. So the cost can still be reasonable despite low volumes.
Apparently, the bean counters do not agree with your Nobel level analysis. While it may be true that a plain vanilla reference x86 platform is easier, you forget that we do not make plain vanilla stuff here.
The organization as it's currently constructed seems to have more people than it needs to design and support products for the cloud groups.
@srm: you are missing my point. The original question was -
Why invest in R & D, manufacturing and supply chain management for such low volume products ?
And the answer is that Oracle doesn't have to do the manufacturing and SCM parts. Besides, these days it doesn't take a lot of R&D to design an x86 box. So the cost can still be reasonable despite low volumes.
Ridiculous MSRPs of Oracle x86 servers do not reflect their cost to the company.
Since when do MSRPs reflect the cost of anything to the company that produced it?
As far as I know, none of the current x86 server line was intended to serve the bottom of the barrel commodity server market but rather for engineered systems.
Oracle Cloud is certainly not paying MSRP for the servers they are deploying.
Of course not. Did you pay MSRP for your car?
Ridiculous MSRPs of Oracle x86 servers do not reflect their cost to the company. AFAIK, X6 and before were made by Mitac and X7 and on are made by Foxconn and their cost is in line with other similar hardware manufactured in TW or China. Oracle Cloud is certainly not paying MSRP for the servers they are deploying.
I haven't seen any evidence of the cloud teams providing significant guidance on features.
That's because you are non involved with these teams on a day to day basis. There's plenty of guidance on feature sets coming from the could teams.
The only justification I saw on other threads on this site was that the folks that run the Oracle data centers see value in having a team internal to Oracle that can support these systems, but I question whether this justifies the investment.
Well, ask yourself this question, if you buy from a commodity vendor, how much support, customization and hand holding are you going to get? A lot of the "value" you are seeking is hidden underneath those things and by the time you negotiate it back, it's at best a wash or way more than what O's currently paying for it's internal teams. You think that the bean counters would pissssssss away $$$ if they can have it any other way?
This thread was originally about whether it makes sense for Oracle to continue to develop their own x86 systems, but it's taken a turn towards x86 vs SPARC. I'd like to get back to original question by the OP.
It doesn't seem like it makes sense for Oracle to continue down this path. I haven't seen any evidence of the cloud teams providing significant guidance on features. The only justification I saw on other threads on this site was that the folks that run the Oracle data centers see value in having a team internal to Oracle that can support these systems, but I question whether this justifies the investment.
This is "-cku" again.
Just to provide a comparison with X5-2s... We have not had a single MB failure on any of our ~20 X5-2's during the same period. That's why we have moved ahead with X6-2's (instead of SPARC i.e t7-2s).
Question was not SPARC v/s X86
Maybe you should read all the replies in this thread. The very first reply to this thread was a typical SPARC fanboi blurt.
simply outsourcing this to OEM and save $$$$
How do you know outsourcing to OEM will save $$$?
If I were Dell or HP I would make certain that I treat Oracle as customer the same exact way Oracle treats its own customers.
Karma's a beach.
@-jim
Clearly you have not come across (used) a recent SPARC based machine it seems...
We have been using T5-2s (about 30 of them) for the last 2.5 years, with Sol 11.3. I must say that (as it was my first experience with a non-x86 server machine) I was totally disappointed with Solaris/SPARC. I do appreciate some stuff in Sol (which is much ahead of Linux ATM).
We had to replace all motherboards (on some machines multiple times) on all of the 30 machines. PSU's were also replaced on all 30 machines. This all happened after we were hit by multiple downtimes. These were bcuz of HW failures (where a component such as a DCDC converter, USB "bridge" b/w MB and ILOM SP failing, etc) MB failure suddenly not booting) and Sol 11.3 bugs.
Most of these OS bugs were ZFS and memory subsystem related.
I was able to find out 2 bugs in 3 days in some very commonly used utilities [am not proving names here]. One of the regressions even hit a recent CPU of Sol 10 [or may be it was present forever].
Oracle offered us T7's but we made sure to stay clear of those.
I ma not saying ex-SUN HW/SW designers are not good, may be they didn't have enough time, and their products were launched w/o due testing.
"Given enough eye balls, are bugs are shallow" --- No one can beat that.
Question was not SPARC v/s X86
Question was does it make sense for Oracle to design and build low volume product like X-7x.
They employ hundreds of employees in engineering and WW ops org to release and sustain these products through life cycle v/s simply our sourcing this to OEM and save $$$$
Performance per thread: Sparc M8 doesnt beat Intel x86 Haswell.
Here are some numbers for SPECint. Admittedly, they are pretty old - from 2010. 2010 seems to be the last year that Solaris SPARC has published its numbers for SPECint.
Solaris SPARC system: SPARC Enterprise M8000, SPARC64 VII+, 64 cores, 16 chips, 4 cores/chip, 2 threads/core, 3.0 GHz. SPEC submission date: October 2010. Running Solaris 10 09/1
https://www.spec.org/cpu2006/results/res2010q4/cpu2006-20101206-13888.html
Intel Xeon system: ASUS TS500-E6 (Z8NA-D6) server system, Intel Xeon X5670, 12 cores, 2 chips, 6 cores/chip, 2 threads/core, 2.9GHz. SPEC submission date: July 2010. Running SUSE Linux Enterprise Server 11:
https://www.spec.org/cpu2006/results/res2010q3/cpu2006-20100730-12706.html
There are two columns in the SPEC Results Table: Seconds and Ratio.
Seconds is the clock time taken by each benchmark to run.
Ratio is the placement of each benchmark on the totem pole of benchmark speeds.
Seconds: the lower, the better.
Ratio: the higher, the better.
Each benchmark was run three times, for two different SPEC configurations: Base and Peak.
How can Solaris on SPARC compete with Linux on Intel Xeon?
I realize that SPARC CPU's have evolved since 2010. So have Intel CPU's.
It seems a poster above is a big fan of Sparc/Solaris...and hasn't looked at X86_64 in YEARS. What was once the IBM X3950...and is now the Lenovo X3950 absolutely blows away anything Sparc/Solaris...when the modules of an X3950 are put together...you can have a system with 172 cores, mirrored RAM and you can find software for it...because it runs Linux. That is 172 X86_64 cores. You want to talk about a machine for virtualization?
Performance per thread: Sparc M8 doesnt beat Intel x86 Haswell. Threads on my $300 i7 6700 is faster by some margin than all the M8 I could have tried with real benchmarks : compilation, crypto, you name it.
Sparc has many more threads though, but that alone is not enough to justify the price tag.
I can't say about M9 and nobody will ever.
It's a wise decision to kill Sparc because at the end of the day it's just an ARM on steriods. And ARM is low-power, that counts in a multi-thousand CPU datacenter and they will come eventually to hundreds of threads per core.
It's just a matter of time before SC and MH comes out with their spreadsheet and move the cloud hardware to be manufactured by Taiwanese vendors. It's very expensive for Oracle to manufacture x86 systems either for cloud customers or for customers.
Agree that SPARC is dead. But why design and build your own x86 server when you can buy same at half the cost from OEM. ?
Nobody uses SPARC/Solaris anymore as it's too expensive. Nobody has SPARC/Solaris skill as all UNIX style development is done via Linux/ARM or Linux/x86. Nobody has the skills to fine tune a SPARC/Solaris box to get the most efficient configuration out of it. They don't want to hire Oracle Professional Services as that means more vendor lock in. Companies are not interested in an eco-system which has not got their best interests in mind and has limited technical staff in the market who have to support it. Solaris lost the mindset of most developers 10-15 years ago and that's the biggest problem that you don't have a developer base to evangelize the platform. Solaris was great for its day but this is a Linux world now.
Good luck!
That's a very good Q.. likely that Oracle needs to build them into their very nice (but expensive) Engineered Systems. The funny thing is that it's SPARC/Solaris products are Soooo much faster and better than any other servers (and OS) on the market (pretty much all world records for any Oracle products), and that Oracle is afraid to market the benefits for sake of competing with it's own x86/Linux products to show those benfits publicly and upset its internal x86/Linux Executive lemmings or those that still are followers of other old converged infrastructure, notably the blind x86/Linux crowd (who use primarily very CRAPPY throw away servers and an OS with NO intrinsic built in security nor alternate boot environments, that can't scale linearly nor vertically, nor use memory pages beyond ~3MB, where SPARC/Solaris offer 64 GB Memory pages !). That doesn't included all the secret DB, Java, and encryption co-processors + the only CPU secured memory (that stops hacks). Word on the street is that Oracle is about to release new SPARC CPU's and servers that are 2-3x faster PER CORE than x86 ! .. so the consolidation and reduced licensing savings should be HUGE (+ their free Virtualization and mgmt tools with support). Why my company is blindly following the Linux/x86 lemmings is beyond me where we take several hours to days to diagnose issues with several vendors, compared to our 5 year+ old SPARC/Solaris boxes that literally need to have their cords pulled to make them hiccup. Who knows, but old habits and sticking with BAD corporate strategy is easier than admitting to shifting past decisions and egg-on-face.