Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 10 Nov 2015 22:55:38 -0800
From:      Jordan Hubbard <jkh@mail.turbofuzz.com>
To:        Warner Losh <imp@bsdimp.com>
Cc:        Anna Wilcox <AWilcox@wilcox-tech.com>, freebsd-arch <freebsd-arch@freebsd.org>, Sean Bruno <sbruno@freebsd.org>, sparc64@freebsd.org, Marius Strobl <marius@alchemy.franken.de>
Subject:   Re: Sparc64 doesn't care about you, and you shouldn't care about Sparc64
Message-ID:  <39947478-4710-47D8-BAB1-FC93979570B6@mail.turbofuzz.com>
In-Reply-To: <CANCZdfqO-SdjnonGzRr2H0pDon5oALsDGsmG3KOxPGRVdTbHPQ@mail.gmail.com>
References:  <563A5893.1030607@freebsd.org> <2AAC0EF3-528B-476F-BA9C-CDC3004465D0@bsdimp.com> <20151108155501.GA1901@alchemy.franken.de> <563F8385.3090603@freebsd.org> <56417100.5050600@Wilcox-Tech.com> <CANCZdfqO-SdjnonGzRr2H0pDon5oALsDGsmG3KOxPGRVdTbHPQ@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help

> On Nov 10, 2015, at 9:54 PM, Warner Losh <imp@bsdimp.com> wrote:
>=20
> sparc64 is the odd-man out currently. However, even if clang doesn't
> work, the gcc external toolchain works well for other platforms. If it =
makes
> sprac64 more viable, then so much the better.=20

Hi Warner,

I hate to be a voice of pragmatism here when we=E2=80=99re having so =
much fun discussing it from an architectural perspective, but=E2=80=A6

What=E2=80=99s the actual goal (from a future market relevance =
perspective) of putting resources, any resources, into sparc64?  I think =
that=E2=80=99s the key question that needs to get asked and answered =
here since we all know that:

1) FreeBSD is not NetBSD - it has never historically supported =E2=80=9Cx8=
6 alternative architectures=E2=80=9D just because they existed and might =
be technically interesting to port to, there had to be some sort of user =
community numbers to justify the time and energy expended for the =
project as a whole (and even in an all-volunteer driven project, there =
is simply no such thing as =E2=80=9Cfree=E2=80=9D - everything has a =
cost somewhere).

As phk noted earlier in the thread, the ALPHA port was an exception to =
this rule simply because it was the first-ever 64 bit port for FreeBSD =
and we knew it would buy us some much-needed 64 bit cleanliness, but it =
also fell off the support roadmap and into the history books once =
ALPHA=E2=80=99s market relevance had clearly ended.

NetBSD/alpha still exists, all the way up to and including NetBSD 7.0, =
because their slogan is =E2=80=9COf course it runs NetBSD.=E2=80=9D   =
Again, FreeBSD !=3D NetBSD.  The emphasis on market share is and always =
has been a key differentiator for FreeBSD and part of both its own =
slogans and mission statement.

2) Sparc64 global market share has declined significantly since Oracle =
purchased Sun, leaving Oracle and Fujitsu as the only two significant =
players in that market.  Sure, putting =E2=80=9Cold equipment to work=E2=80=
=9D is also always a tempting objective, but it=E2=80=99s one that =
really requires viewing through an objective lens since the perspective =
of someone who owns said "old equipment" is rather more biased than the =
perspective of the market as a whole.  The market as a whole appears to =
consist (in terms of global server market share):

HP (x64)			27.6%
IBM	(x64)		22.9%
DELL (x64)		16.4%
All others (x64):	24% (combined estimate, including Cisco and =
Huawei)
Total:			90.9%

[ Source: Gartner ]

That leaves 9.1% for the rest of the server industry, which includes =
Itanium, POWER4 and SPARC64.   We can also probably safely assume that =
even amongst that tiny 9% pie slice, vendors are focused on the future =
since their overall market share is declining (about 5% annually), which =
begs the question:  Is FreeBSD/SPARC64 aiming at the T5, even while =
Oracle themselves are shifting emphasis to lower-cost x64 systems for =
which FreeBSD is already competitive, or is it really just trying to =
keep some older collection of increasingly power/performance inefficient =
(by comparison) alive?

Again, what=E2=80=99s the long-term goal of supporting this =
architecture?  The old adage about =E2=80=9Cpicking your battles=E2=80=9D =
applies here, no matter how enthusiastic the small community of =
remaining SPARC users might be, which is why I am risking lightning =
bolts of wrath from SPARC zealots in even daring to ask the question. =
:-)

Thanks,

- Jordan




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?39947478-4710-47D8-BAB1-FC93979570B6>