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

[Stephen Hoffman]

On 2015-03-11 15:37:02 +0000, <lists at openmailbox.org> said:

> On Wed, 11 Mar 2015 10:27:48 -0400 Stephen Hoffman via Info-vax 
> <info-vax at rbnsn.com> wrote:
> 
>> What's your alternative to x86-64 here?   How much will the "special  
>> kick booty chip" cost to produce and integrate?
> 
> There's a lot more wrong with Intel than can ever be fixed.

Undoubtedly.  But that's not a dragon that VSI can slay.

In short, what's your alternative here?  SPARC just isn't it.

Whether ARM — which has its own warts, not the least of which is the 
confusion and the misuse of TrustZone, and that SBSA is only just 
getting going — will see increasing success in the server space, or 
whether the Mill CPU can get traction?

Right now, it's Intel (and AMD) x86-64, and ARM as the most 
likely-looking new candidate being ARMv8-A and SBSA designs.

> I agree a good processor would be nice from a programmer's POV and from 
> a healthy market POV.

Good and elegant processors can and often do fail in the market, 
whether we're discussing VAX or Alpha or other designs.

> SPARC is an open design. I don't know who's fabbing them today but if 
> it isn't Oracle itself than there's no R&D cost in using SPARC chips. 
> They're very good and they have some great features but they are not as 
> fast as the fastest Intel chips. If that doesn't matter than SPARC is a 
> great design with a lot going for it.

The great thing about open standards: there are so many to choose from.

Sure, having an open design — as is the case with Open Compute widgets 
I mentioned elsewhere recently — can be a good thing, but you then have 
to have the associated infrastructure and the available products, and 
your products either have to be appreciably better than the competitive 
products, or cheaper, or more efficient.

Right now, not running on the mainstream Windows x86-64 platform means 
that whatever processor parts and server parts and system parts you 
have is selling in lower volume, which means your prices are inherently 
going to be somewhere between higher and much higher, which means you 
inherently have a deeper competitive hole to dig yourself out of.

Whether OpenRISC <http://en.wikipedia.org/wiki/OpenRISC?> or some other 
open-source hardware 
<http://en.wikipedia.org/wiki/Open-source_hardware> gets you there, 
eventually?  That depends on how many designs and how many systems 
become available, and particularly whether — and as end-user folks have 
only rarely been willing to do — whether enough folks are willing to 
migrate to what is an incompatible hardware to really matter; whether 
there's enough interest to get the volumes high enough, and the prices 
down.

> 
>> 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 has been some discussion on various Solaris mailing lists about 
> all this. What you wrote is entirely intentional on Oracle's part. I 
> would seem there is great ego involved and getting the bish fish with 
> killer deals and ignoring small-mid sized companies is what it is all 
> about now.

I'd wager that Oracle sales would be happy to sell to smaller firms, 
but they're running into some stiff competition with the likes of 
PostgreSQL, SQLite and other databases.  Some firms can and will need 
vendor support, and that might mean commercial support for their 
current databases, or a migration to Oracle.  But there's a whole lot 
of SQL code — and more than a little NoSQL code, for that matter — 
which does not involve Oracle.  Oracle wants to change that, but 
whether they can do that often enough by moving folks up to their 
premier products from MySQL remains to be determined.

> Oracle no longer markets or views SPARC as a general platform, it's 
> only used as an appliance engine for EXA-boxes (Oracle database 
> appliances, etc.) Solaris will eventually probably go away because of 
> this "strategy."
> 
> A shame, because Solaris is a nice OS and the servers are very well 
> thought out and well made. OpenBoot is great. SPARC is a great 
> platform.  And Solaris really does run better on it than on X86. You 
> can feel it.

Oracle is undoubtedly looking at their costs and their revenues, and 
particularly at the associated trends.

Being on the wrong end of a market commoditization is not a pretty 
place for any business, as you either need to drive massive volumes and 
drive your costs down (q.v. 
<http://en.wikipedia.org/wiki/Economies_of_scale>), or you need to have 
enough of a value-add that folks still want to buy your gear.  
Everybody's selling generic x86-64 boxes running Windows, so you're not 
going to have an easy path to that, other than through cost reductions. 
 The folks that are shipping generic x86-64 boxes running Windows are 
basically trapped by each other and by their ever-dropping prices they 
can sell their equipment at — for pricing, there's 
<http://gizmodo.com/hp-stream-11-review-200-and-worth-every-penny-1668809067> 
and 
<http://arstechnica.com/gadgets/2015/02/cheap-functional-upgradeable-hps-stream-and-pavilion-mini-desktops-reviewed/> 
— and thus the vendors are left to make revenue with add-ons (and that 
occasionally cause problems 
<http://arstechnica.com/security/2015/02/ssl-busting-code-that-threatened-lenovo-users-found-in-a-dozen-more-apps/>) 
or through post-sales services and support and add-ons — and HP is one 
of the few vendors that is reasonably well positioned here and 
particularly for the business markets, or the vendor can try to add 
value — which takes time and effort and with no clear and no certain 
path to a return even assuming the cash is available for that effort, 
and whether it's trying to create value from custom software or a 
custom OS or custom add-on features such as distributed management — or 
to consolidate and buy up competitors, or the vendor can sell off and 
exit the market while the current products and services and retail 
shelf space and trademarks still have some value to potential buyers.

Elegance and simplicity can sell, if you invest in what that costs, 
really work at all of what that means, and if you can market it.  
Effectively.   But an elegant processor architecture?  That's a much 
tougher sale to make to an end-user, particularly given how much more 
that elegant processor can cost as compared with a design that's in 
ginormous production volumes.

FWIW and if processor elegance is of general interest to you, maybe 
have a look at the ARMv8-vintage architecture, within the current crop 
of potential choices.



-- 
Pure Personal Opinion | HoffmanLabs LLC