IBM recently
announced they will open the POWER architecture to align with open software
rules. Further, IBM will license its POWER ISA[1]
to the OpenPOWER Foundation to enable royalty-free implementations on POWER. This
is a big deal. First, a comment on an IBM perspective and a journalistic view.
Then, our thoughts.
OpenPOWER GM
Ken King’s blog[2] provides his perspective on the
announcement’s significance. As you will see, we generally agree with him. But
we take a wider view as we consider some risks in IBM’s plan. An article[3] in DataCenterDynamics, discussing
the licensing focuses heavily on this move as a reaction to the end of Moore’s
law. This is only partially true. IBM has aggressively promoted (and pursued)
the idea that the focus of innovation must shift from the CPU chip to the
overall system. Given that the two most powerful supercomputers in the world
are POWER9 based, IBM’s strategy is valid and should continue; opening POWER
will only help the process.
First, with
this move POWER’s instruction set details become completely transparent rights
to implement open to anyone. Developers can view instruction specifications and
have the right to develop new hardware using POWER’s instruction set.
Developers can enhance new versions of POWER systems with patentable
innovations. While backward POWER compatibility will be protected, other compatible
changes will require majority vote approval by OpenPOWER group members. However,
IBM can (and will) continue to enhance the ISA in the AIX compliance subset and
with optional features. Note that IBM is not giving over complete control of the POWER
architecture. Expect new versions of POWER to appear over time; just as new
Linux versions do.
Next, there
are financial implications for IBM (revenue impact, net investment recovery),
these lie outside the scope of our research. There are international
implications to consider. There are organizational and governmental concerns
over risks of embedded malignant code and/or back-door access points, as well
as uncertainties in renegotiations to trade treaties and relationships. Both entities have become very sensitive to potential (backdoor)
openings in technology that could be used for espionage purposes. OpenPOWER has
obvious appeal in such environments. Also, for example, a looser connection to IBM may add
significantly to the platform’s existing international appeal. For instance in China,
where IBM has significant business and research investments involving multiple
products.
There exists
yet another potentially connected development.
IBM recently completed their Red Hat purchase. Red Hat developed a
highly successful business around open Linux software. It isn’t much of a
stretch to imagine IBM hoping to realize similar success built around open
hardware.
However, hardware
and software markets differ in significant ways. The software market is far
larger than that for hardware. Customer buying patterns are totally different.
Hardware profit margins are thinner than software’s. The competitive landscape
differs significantly. Different barriers-to-entry exist in the hardware
business than for software. Finally, there are far fewer hardware engineers
than software programmers; the talent pool available to develop and evolve POWER
is much smaller than what Linux draws upon.
Considering
these factors, IBM’s strategy faces some significant risks. The POWER
acceptance that IBM is looking for (and needs) may not exist. It took time to
reach today’s wide market acceptance of Linux. While popular, POWER hardware
may not have the same ability to grow demand. Other companies may not see the
value in developing POWER systems to satisfy a not-so-dynamic demand.
Clearly, HP
and Dell never felt compelled to offer a POWER-based system; correctly viewing
it a direct competitor to their existing offerings. Nor are they “married” to
any chip; each offers both Intel- and AMD-based systems. We think they will
closely monitor POWER growth. They would need a very compelling
market/technological reason to add a POWER-based product. At present, there
appears no rush to do so, but that can change. However, even if both companies
offer POWER products, it would not necessarily guarantee long-term success.
Before continuing, IBM explicitly targets OpenPower at
the hi-performance, domain specific and heterogeneous computing, not the
generic market. IBM describes it this way “open strategy is focused on
capturing the future white space and growth in purpose-built computing/domain
specific computing/heterogeneous computing.
Of course, we may get legacy too, but the market opportunity is in all
the things not built yet that people are looking to build". They have had
some significant success there. The risk lies in the rate of demand growth along
with the number and entry speed of competitors pursuing that market.
There is
speculation that this move provides IBM a graceful way to exit the POWER
business. We cannot prove this view is wrong. But we think it highly unlikely
for the following reasons. Current IBM POWER customers fall into one of four
categories: Linux, AIX, IBM i and supercomputer customers. AIX and IBM i are
legacy customers. The remaining two are in dynamically high-growth areas.
Supercomputers were a highly promoted, specifically targeted segment. Linux is
an all-in market with the Red Hat purchase. Today, taken together, they
represent a significant portion of IBM’s revenue. IBM exiting or dead ending
some or all, would be embarrassingly bad PR and a financial blow. We reiterate,
possible, but highly unlikely to happen.
IBM may be
planning to reduce overall POWER investment by passing future enhancements to
OpenPOWER. However, we don’t find this believable. How negatively this might
affect current customers depends on how aggressively OpenPOWER (and other
vendors) invest. This, in turn, depends on OpenPOWER’s market appeal. The
amount and timing of on-going IBM financial and technology support for
OpenPOWER is equally influential. We reject this idea. We do not think IBM has any plans to reduce investment. IBM
retains the option to enhance the ISA in the AIX compliance subset and add
optional features. This ensures IBM can continue to enhance the ISA.
We think
current POWER customers (whether Linux, AIX or IBM i) have little short- or
medium-term downside risk. They can reasonably expect continued innovation and support.
Some hardware firms have already voiced support of IBM’s move. In fact, if new
POWER-oriented companies keep entering the market, customer hardware choices
may increase. The Power ecosystem have interest to further enhance
accelerators, memory controllers, systems, etc.
IBM i
customers using IBM’s proprietary OS are in a unique position. Moving workloads
to another OS is expensive. They must rely on IBM for new device and technology
support; a strategy that has historically worked well. In the longer term, IBM
has a good track record of arranging support for customers when it exits a
business. Note we are NOT suggesting that IBM has any plans to drop IBM i.
Finally,
there is a very good chance that IBM’s gamble on POWER Systems’ future will pay
off. While we are not forecasting runaway success, we expect the number of POWER-supporting
vendors to steadily increase, along with the number of POWER-related products.
Existing users will likely increase their usage, finding new uses as POWER
technology evolves.
On balance,
open POWER will likely benefit customers and vendors. Even those not supporting
POWER will benefit from the competition that will drive responses to future POWER-driven
technological innovation. The new POWER organization will continue innovating.
We will be writing about that as it happens.
[1]
ISA refers to instruction set architecture. It is the terminology used by IBM
and OpenPOWER Foundation.
[2]
Embracing and expanding the open hardware ecosystem, August 21, 2019, https://www.ibm.com/blogs/systems/embracing-and-expanding-the-open-hardware-ecosystem/
[3]
Peter Judge wrote: IBM opens Power architecture to bolster flagging CPUs,
August 22, 2019, https://www.datacenterdynamics.com/news/ibm-opens-power-architecture-bolster-flagging-cpus/
No comments:
Post a Comment