[Info-vax] rx2800i2 sales/support window changes

[Jan-Erik Soderholm]

David Froble wrote 2014-02-15 15:11:
> JF Mezei wrote:
>> On 14-02-14 16:09, Keith Parris wrote:
>>
>>> Customers who have migrated already from (VAX or) Alpha to Integrity,
>>> and traded in all their licenses, old hardware, etc. (or those customers
>>> who started out on Integrity) are unlikely to find moving to Alpha on
>>> emulators attractive at all, for multiple reasons:
>>
>> This is where HP can make things easier.
>>
>>> 1) Performance.
>>
>> At the time Alpha was killed the 8086 hadn't even grown its first chest
>> hair and was nowhere near being mainframe class. So yes, 8086s couldn't
>> give the emulated performance of ALpha because they were slower than
>> Alpha even for native code.
>>
>> Since then though, Alpha plateau-ed and the 8086 matured into a potent
>> 64 bit engine with very good memory interface and surpassed both Alpha
>> and that IA64 thing.
>>
>> And as time progresses the performance advantage will keep on growing.
>> And if emulated Alpha on x86 today doesn't have better performance than
>> a Tukwilla, it likely will do so very soon.
>>
>>> 2) Support. Since HP Mature Product Support for OpenVMS 8.4 on Alpha is
>>> committed only through at least the end of 2018, whereas MPS on OpenVMS
>>> 8.4 on Integrity is committed through at least the end of 2025,
>>
>> See my previous post. HP should change this if faced with reality that
>> Alpha-VMS will outlive (through emulation) IA64-VMS.
>>
>>
>>
>>> I expect the prospect of Itanium emulation to start to become attractive
>>> as individual hardware platforms age and after they lose hardware
>>> support from HP and start to fail frequently,
>>
>> Or when HP hikes maintenance costs for IA64 systems. What is important
>> to remmeber is that HP-UX is the largest user of them IA64 things. So an
>> emulator is likely going to be built for that. This is why the fear that
>> an emulator would emulate Poulson. (especially if Poulson emulation
>> provides greater performance than Tukwila emulation)
>>
>>
>
> It's not clear to me that the above supposition is indeed correct. Hardware
> features are implemented to allow the hardware to perform some tasks
> quicker, sometimes replacing doing them in software.  When emulating an
> instruction set, it's all software.  It's even possible that the software
> emulation of some instructions might be slower than not emulating those
> instructions.
>

I thought the same, it's not at all for sure that a more complex
processor will perform better in emulation. There is just more
to emulate. The fact that a Poulson runs at 2.5 GHz or that the
Poulson has 8 cores has not any rellevance at all. It is the
complexity of the emulation itself that matters.

And about the future perfomance gains. It seems as the processor
industry thinks that the single thread/code performance has more
or less paned out by now.

Performance of an 800 Mhz Alpha on a 3 GHz xeon is one thing.
But it is very unclear if there will be any 9.4 GHz systems
where an 2.5 GHz Poulson could be emulated.

Finaly I think that the cleaner arhitecture of the Alpha is
easier to emulate the the weird instruction format of IA64.



> The idea of emulation is that you can "still do it" vs "can't do it any
> more".  Sort of like old men ....