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

[Stephen Hoffman]

On 2015-03-11 11:06:57 +0000, seasoned_geek said:

> Sadly, putting VMS on x86 will ensure it a market life of minutes, not 
> years. It is a special OS and needs a special kick booty chip. The 
> earlier comments in this thread about Solaris running faster and better 
> on SPARC T5 should provide a guiding light, but it won't. By all 
> accounts Solaris should have been long gone from the market by now, 
> but, a certain segment of the market cannot make do with Wal-mart 
> quality chips so Solaris continues.
> 
> VMS also serves a market segment which cannot make do with Wal-mart 
> quality chips. Porting VMS to Wal-mart quality chips ensures its 
> demise. Porting VMS to the Wal-mart quality x86 line is like feeding 
> chocolate to dogs. They beg for it, but for them it is fatal.

What's your alternative to x86-64 here?   How much will the "special 
kick booty chip" cost to produce and integrate?

Running your own competitive microprocessor designs and your own fab 
requires huge scale and huge funding, or you're off the price and 
performance curve — DEC bailed long ago, HP got Intel to partner and to 
fab their design and that looks to end with Kittson, and IBM is the 
only the most recent entity that is exiting the custom microprocessor 
business.

Alternatives to x86-64?  There's SPARC, and some designs like MIPS that 
aren't as common with general computing, and various ARM designs are 
moving into the low-end.  Or dropping a few billion dollars into your 
own design and your own fab or trying to use an outside fab, and with 
no clear path to break-even.   For example, Intel Fab 42 was to in in 
Chandler AZ, and — though its opening has been postponed indefinitely — 
was expected to cost Intel five billion dollars.

As for SPARC and Solaris, Oracle's business is apparently contracting — 
they're profitable and variously more profitable than they've been, but 
that's over fewer customers, per one of the more recent financial 
reports. There's still good money in those existing Oracle customers, 
but a contracting customer base is not popular with investors, and a 
declining base means application availability for current and potential 
new customers eventually becomes a problem.

For existing Solaris customers that want to move, there's illumos, 
where there really wasn't a good path for OpenVMS customers.

I do still think VSI will be headed toward their own bespoke server 
line directly or through a partnership, even if the folks at VSI might 
not really want to take on that cost and that overhead.  Servers using 
Intel Xeon with ECC memory (q.v. rowhammer), and with decent-grade 
components with available OpenVMS driver support.   Quite possibly with 
some sort of a DSMOSX lock-in for OpenVMS, for that matter.  
<http://en.wikipedia.org/wiki/Apple–Intel_architecture#Dont_Steal_Mac_OS_X.kext> 
 Customers probably won't like the ~five to ~seven year server lifetime 
of these servers and of x86-64 servers in general, but then the newer 
servers have been the same or cheaper, and the efficiency and the 
density has been increasing.

Special chips are a differentiator and a value when they're enough 
special to warrant the costs and the hassles involved.  That might be 
faster, or it might mean better battery efficiency, or some other 
factor.   But the cost difference inherent in lower volumes and in the 
added software work means that even chips and systems that are 
ostensibly better designs are a tough sale in the market.  At least 
until you're operating at a scale well past where VSI is right now.

But to your point, whether the x86-64 port works out for VSI, we shall see.



-- 
Pure Personal Opinion | HoffmanLabs LLC