[Info-vax] OT: news from the trenches (re: Solaris)

[Stephen Hoffman]

On 2015-03-12 13:50:25 +0000, <lists at openmailbox.org> said:

> On Wed, 11 Mar 2015 15:43:43 -0400
> Stephen Hoffman via Info-vax <info-vax at rbnsn.com> wrote:
> 
> 
>> "IBM’s POWER chip manufacturing volume isn’t enough to sustain the  
>> business on its own."
> 
> I don't know what IBM is doing and I don't believe that statement. It 
> seems to me some people who really don't understand what IBM has 
> traditionally been about, or perhaps they understand and don't care, 
> are making some decisions that are going to come back to haunt them in 
> the not too distant future. IBM has always been a hardware company at 
> the core. They write some damn good software and provide services but 
> that's to sell hardware.
> 
> It's like they threw out the baby with the bathwater.

The folks at IBM are making decisions based on their best available 
data, including on the revenue trends and available investments.

Hardware and associated production capabilities that are now or that 
are becoming commodities do not look to be the future of IBM, either.

Keeping a fab filled is how you run those hugely-expensive things most 
efficiently, and if you don't have enough of your own product volume to 
keep the fab running, you can either try to be a foundry for others and 
as Intel is now trying and IBM once did — IBM fabbed EV68C and EV68CB 
Alpha chips for DEC — or your prices for running the fab and for your 
chips increase to untenable levels and you're left with no path other 
than the exit.

As for the future, IBM does have a deal for iOS devices in the 
enterprise, which is an interesting juxtaposition with past history.  
But I digress.

>> IBM has sold off their x86 business and most recently their x86 server 
>> business to Lenovo, and I'd wonder of POWER isn't following — albeit 
>> following along the ten-year roadmap that is available.
> 
> As you know IBM didn't design x86 and it was never their baby.

IBM did the system design and the specs, and used a processor and 
software from elsewhere.

IBM made the PC acceptable to business and from there to many 
consumers, and IBM sold more than a few of their PC and PS/2 boxes.

Whether it was "their baby" or not, the PC was synonymous with IBM for 
many years.  The whole business market was PC-DOS, and only later 
MS-DOS or Microsoft or Microsoft Windows.  Once IBM waded in, everybody 
else ended up needing to be compatible, or out of business.   There 
where the legal squabbles around reverse-engineering the BIOS, and the 
PC-compatible market.  Early players there were Phoenix Technologies 
for their BIOS, and Compaq for their Portable.

But again, this old IBM PC stuff is entirely the wrong direction — this 
is about what's happening right now, and what's going to happen in the 
next three or five years or so; however long it might take you to get 
your OpenVMS port completed and out to market, and to start to 
establish a base of users.  IBM and Oracle — and VSI, for that matter — 
have to look ahead.

> That they sold it off makes perfect sense and doesn't tell you anything 
> about their historical major product lines where they controlled the 
> ecosystem end to end. That's what IBM does. They have a lot of bits and 
> pieces but they really shine when it comes to designing hardware and 
> peripherals and an OS to run specifically on those architectures. 
> Selling off z/Arch and POWER fab is a major directional change and 
> based on that I would bet money one of two things is going to happen. 
> Either the leadership is going to realize what they've done and bring 
> back the fab to corp. or IBM will be a much different, smaller, and 
> less profitable company then we have known.

IBM is not going to reacquire the fabs, and they're not going to get 
back into either the x86 or the fab business, and it would not surprise 
me to see IBM eventually migrate to commodity x86-64 processors even on 
their mainframes, assuming they don't spin off the z business.

> POWER is open so as long as somebody is fabbing it anybody can make a 
> server. If they quit making it then yes, there would be a problem. 
> Could that happen? Certainly.

Lower volume means fewer choices and higher prices, and with some 
degree of risk of having the supply vaporize, too.

> 
>> Put another way, if I was to be beholden to any particular vendor(s) 
>> for my platform hardware beyond the Hewlett-Packard Enterprise folks  
>> after HP splits, right now that's probably going to best be Intel  
>> and/or AMD, and the various Windows-box providers, as they have the  
>> highest volumes and the lowest prices.   This if I weren't going to be  
>> using one of those providers to create a bespoke x86-64 box, maybe with 
>> a "DSMOSX-like" feature, specifically for OpenVMS, and that can also  
>> run Windows and/or VMware and/or RHEL and/or other x86-64 software.
> 
> Yeah but you seem to have suggested there's a question whether VMS can 
> or should run on cheap hardware given the VMS customer is typically a 
> high end customer.

You're suggesting that x86-64 isn't high-end, though HP is working to 
provide that with ProLiant Gen9 and newer, and with boxes such as 
Superdome X DragonHawk.  Or with their HydraLynx box, though HP has 
been rather quiet about that box lately.  Maybe waiting for Skylake?   
Not that all of the major hardware vendors are working on designs and 
products for Intel Skylake, though.

> Wouldn't it make sense for a port to a premium hardware platform, as 
> long as it was open and still being fabbed, as opposed to a crapware 
> hardware platform that can run "free software" and Windows?

If Oracle and IBM aren't doing what makes the most financial sense, 
then sure.  Pick a (very nice) processor that comparatively few folks 
are using, and where that design is in direct competition with Intel 
and x86-64.  But I wouldn't.   If you want to go with what is, there's 
already Itanium, and with the Kittson-class servers when (if?) those 
arrive.

> If you think about it I'm not sure porting VMS to Intel is not going to 
> be suicidal. After they see they can run on cheaper hardware, isn't 
> cheaper software the next step? And Linux is free and eventually some 
> pinhead is not going to approve the budget for that wierd OS nobody 
> ever heard of, since Linux is free...

Have you ported OpenVMS applications to Unix?   It's not the processor 
that's the source of all the effort.  The underlying processor really 
isn't even a rounding-error-level factor in the effort involved in most 
ports, and there's very little or no processor-specific assembler code 
in most applications.  The costs and the effort involve all the OpenVMS 
APIs and dependencies and layered products that need to be ripped out 
and replaced.  VSI's entire current business model is arguably based on 
that porting cost, too.

As for x86-64, access to x86-64 tools such as VMs is useful, and the 
ability to have backplane-speed access to Windows or Linux can be handy 
for some uses.  HP created their own VM with Integrity servers.   That 
work won't be necessary with x86-64.

> 
>> Ask yourself what architecture you'd pick here after Itanium, if you 
>> were spending, say, ten million dollars of your own money, and wanted  
>> to maximize your chances of success when you were launching an  
>> operating system business.  Sure, if you had infinite money and/or 
>> infinite staff, x86-64 and POWER and MIPS and ARM and SPARC and  
>> Ooooh-Shiny and whatever else can pass the Turing test...  But if you  
>> had to pick just one, and bet on it?
> 
> I would say POWER but it remains to be seen what the future holds in 
> terms of chip availability and development etc.

None of us can know that future, and when you're responsible for making 
these decisions, you gather the available information and make the best 
decisions you can.  Once you start to invest in an operating system 
port, everything is a sunk cost, and if you guess wrong, you've lost 
all the associated time and effort and investment, and the schedule 
time involved too, as everybody else is making forward progress in 
parallel with your porting effort, and any customers that couldn't or 
wouldn't wait for the new platform.

The only bet against Intel and high-volume x86-64 I can see here is 
with ARM, and the various producers are not quite up into the 
availability and configuration and performance range  that OpenVMS 
customers seem to want.  Yet.  I'd want to see more standardization 
across at least some of the available ARM designs, so that everybody's 
not chasing specific cores; this whether SBSA or otherwise.  As POWER, 
IBM looks to be stepping away from that — even with the ten-year 
roadmap —


-- 
Pure Personal Opinion | HoffmanLabs LLC