21 Jan
2023
21 Jan
'23
1:08 a.m.
On Fri, Jan 20, 2023 at 11:39 AM Theodore Ts'o <tytso(a)mit.edu> wrote:
On Wed, Jan 18, 2023 at 05:09:39PM -0800, Larry McVoy wrote:
From my understanding, Sun was up for sale because of competitive
challenges with the high-end servers (due to delays in their high-end
Sparc chips, such as Rock) against products such as IBM's Power
(pSeries) machines. These systems had a much better margin, and so if
you're making money primarily off of hardware, this segment is super
important. The x86 servers don't make as much money, which is why IBM
would end up divesting their xSeries business to Lenovo.
IBM was primarily interested in Sun for the Java business; it was
super important for IBM Software side of the business, since all of
its major products (Webshere, Tivoli, etc.) were written in Java. IBM
didn't really care about Solaris or the Sparc business; after all,
IBM's pSeries with AIX was doing quite well from a sales perspective
in the customer segments that were most important for IBM.
When I was part of the IBM Linux Technology Center, I participated in
an IBM-wide study about whether or not it made sense to invest in file
system technologies. What was interesting about it was that it was
*purely* from a business perspective; would it drive business to IBM?
Would IBM customers find it useful enough to pay $$$ for it? IBM's
decision to not try to invest in some of the cool technologies like
those that ZFS was pioneering was purely made as a purely cold-hearted
business decision. Whether it was cool technology or not didn't enter
into the evaluation and decision function.
GPFS (now called Spectrum Scale?) is one of the most featured
filesystems I can think of, I bet it met both definitions?
I was the
person nominally in charge of the OpenSolaris port to z (Neale
Ferguson did most of the heavy lifting) when Sine Nomine built it, having
read the tea leaves and believing that IBM would buy Sun. And then IBM
tightened the screws a little too far and Larry Ellison grabbed it
instead. Dammit.
Yeah, I'm not a Solaris fan (because SunOS) but there was some good
technology in there. Would have been cool if IBM kept it going. I
never really understood why Sun was up for sale. I'm not going to say that this way of making
technology decisions is
perfect; it definitely has downsides. But I *am* sure it reflected
what IBM was willing to pay for Sun Microsystems the company --- and
Sun was hoping for more $$$ for its shareholders, which is a
completely fair attitude. Was Sun Microsystems worth more to Oracle?
I'm not sure, especially since Oracle has mostly treated Solaris as a
program loader for Oracle Enterprise Database. But at the end of the
day Larrison Ellison was willing to pay more, whether or not it was a
principled business decision, or just a desire to take home the Sun
Microsystems trophy. And at some level, it really doesn't matter.
Realistically, I'm not sure Solaris would have fared that much better
under IBM's stewardship. I'm sure IBM would make Solaris available to
those customers who wanted to use it, and IBM would have maintained
Open Solaris as a open source project. But the decision on how much
to invest into new technologies like DTrace and ZFS would have been
made the same way that IBM *declined* to try to create a next
generation file system for AIX or Linux. And the DTrace and ZFS
technologies would have been integrated into Linux (under the GPL
license) and AIX, thus adding Solaris technological distinctiveness to
those OS's. And while Sun's existing customers might still want
Solaris, IBM's customers would very likely stick with the AIX and
Linux that they knew.
One bit of anecdote, AIX does have a dynamic tracing system comprable
to DTrace called ProbeVue. So I think the relative cost of new
engineering on whatever Sun had to offer was never too high. And that
is congruent with my understanding of the technology business, most
times acquisitions are to gain customers (aka revenue) and seldom
employees (aka aquihire). Business technology transfers are usually a
sign of something gone wrong, with occasional flukes of success.
So that would leave Open Solaris competing with Linux
as an open
source project, without necessarily IBM investing much into Open
Solaris except from a hardware enablement perspective, and with the
best Solaris features getting cherry-picked into Linux. So it would
ultimately depend on how much external investment from other companies
might make into Open Solaris versus Linux. And there, a lot of Linux
investment came because its use in the embedded and mobile space.
(Linux's ext4 encryption and fsverity features was for Android and
ChromeOS; it was *not* developed for the data center use cases,
although there are now some use cases starting to pick up the data
center world.) Would Open Solaris been flexible enough to fit on
wrist watches and handheld phones? It's definitely an interesting
question, especially, given Linux would have a head start in those
worlds.
- Ted