Re: AMD vs Xeon 64 bit

Jurijs,

I think it is CYA by the decision makers. Like in the Mainframe days: if you chose IBM you couldn't loose your job (like now with choosing Oracle as the DB).

The customer I dealt with, on two replacement projects, first of all couldn't believe that HW so cheap was able to perform so well. Then all the usual arguments: "we have standardized on Solaris", "we want UNIX", "Not all tools are supported on this new platform", "we don't have skills for this". Finally business sense starts to take over. When the customer wanted to test the machines, SUN couldn't deliver them for a while ;-) The savings are tremendous, and instead of doing maintenance contracts, you just buy a spare machine ;-) One of the things that SUN is doing is talking about how great Solaris is. Solaris is great but that is only important to the IT department not to the end users (business sense?)

My take on the RISC - x86 thing is that AMD is pushing Intel very hard at the moment. Even Dell will release AMD servers by the end of this year. AMD has currently great server chips (the best price/performance/watt for the Opteron) , but I have no doubt that Intel will produce great server chips in 3 years time. James Morle made the comment that you want computers made in the consumer product factories. The reason is that the specialized servers (SUN/IBM/HP) need special manufacturing skills/lines and things will always break (Murphy). So you pay twice. Once for hardware that is more expensive because it is so special and because it is so special it will break often. Mass produced hardware may not be so special but that is the great thing about it, it was mass produced and has better QA (because there are so many customers :-)) it won't break so often. So where are the mass produced (I mean like consumer PCs) RISC computers?

Will RISC beat x86? Sure, but what is the price that you want to pay for this?

I have been toying with two new acronyms: DCP and FBI Disposable Computing Platform
The hardware is so cheap that it doesn't matter any more

Failure Based IT
I think that one of the driving design motives has been to make sure that hardware and software don't fail (RAC anyone?), but I think that this is the wrong approach (I will take some heat for this, I know). A better approach would be to understand that hardware and software will fail and that there is nothing that you can do about this. Companies like Google, Amazon and Yahoo know this. So if something fails it should fail quickly, so that the rest of the system can detect this. The quicker and cleaner something fails the better that is. So that is why I think that most systems don't need RAC but can live with standby databases or shared disk databases.

So with DCP and FBI the hardware really doesn't matter anymore, the only thing is system design (design for failure) is becoming more important.

Anjo

On 9/5/06, Jurijs Velikanovs <j.velikanovs_at_gmail.com> wrote:
>
> Hello there,
>
> I think it is great that nowadays that effective HW exists.
> I tend to recommend x86 platform to my customers as well.
>
> But I can't understood why customers still buying HW based on RISC CPU-s.
> There have to be something good about RISCs beside politics and good
> mng relationship with vendors' representatives.
> Can anybody advice ? (Anjo?)
>
> Please don't tell me that is is about skills that particular customer
> have on board.
> There is no that significant difference between Linux and Unix.
> The savings are dramatic. 1M-30k ~ = 1M
> Even if a customer would invest 10% of savings in people (education) i
> believe the difference will be covered.
>
> Can you advice any field where RISCs will bit x86 ?
>
> Thank you in advance,
> Yury
>
> On 9/5/06, Anjo Kolk <anjo.kolk_at_oraperf.com> wrote:
> >
> > Dl 585 is the one to go for. Buy the fastest and most memory that you
> can
> > and you will be smiling the whole day. I just advised a customer to
> replace
> > their 16 cpu SUN sparc box with a 4 CPU dual core Opteron. The sun box
> was 1
> > million dollar, the opteron costed around 30K dollar. They went live
> over
> > the weekend and and they are very happy :-)
> >
> > Anjo.
> >
> >
> >
> > On 9/5/06, Juan Miranda <j.miranda_at_sermatica.es> wrote:
> > >
> > >
> > >
> > >
> > >
> > > Hi
> > >
> > >
> > >
> > > We will buy an new server.
> > >
> > > SO will be RHEL 4 AS/ES
> > >
> > >
> > >
> > > Alternatives:
> > >
> > >
> > >
> > > 1) HP 585 4x AMD Opteron
> > >
> > > 2) HP 580 4x Xeon
> > >
> > >
> > >
> > >
> > >
> > > Any reference, idea, experience about these 2 alternatives ???
> > >
> > >
> > >
> > > What about SGA limit ?
> > >
> > > I think I can make a huge SGA with AMD64 (no 3 GB limit),
> > >
> > > Do I have this limit in Xeon 64 bit ?
> > >
> > >
> > >
> > > Thanks.
> > >
> > >
> > >
> > >
> > >
> > >
> >
> >
> >
> > --
> > Anjo Kolk
> > Owner and Founder OraPerf Projects
> > tel: +31-577-712000
> > mob: +31-6-55340888
>
>
> --
> Yury
> +44 7738 013090 (GMT)
> ============================================
> http://otn.oracle.com/ocm/jvelikanovs.html
>

-- 
Anjo Kolk
Owner and Founder OraPerf Projects
tel:    +31-577-712000
mob: +31-6-55340888

--
http://www.freelists.org/webpage/oracle-l