[OpenIndiana-discuss] I figured this deserved a separate thread

Tom Kranz tom at siliconbunny.com
Fri Nov 5 10:52:05 UTC 2010


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Paul Johnston wrote:
>
> Please don't take this as trying to cause a flame war :-)
>
> I've never really understood this, is it because of something inherent
> in SPARC (RISC) or just they are built better?
> If you get well made x86 gear will it last as long?
> Or off course is it that the software works better on SPARC as it is
> designed for it and the x86 version is inferior?
Balanced hardware and reliability are the two key factors. The stuff
is built a lot better, so it lasts longer. It's built to higher
tolerances, so it lasts longer, and it's desiged to deal with failure
better, so it can take on higher loads and be pushed more.

Until Nehalem, x86 was crippled with a bus design for shunting data
around. RISC has had the idea of a switch/crossbar/whatever for
donkey's years, and it means reliably I/O transfer rates that can be
sustained. For heavy graphics work, or for thrashing disks or the
network (usually at the same time) this stuff is crucial. It means the
system can scale - in a predictable way. It also means everything is
not CPU-centric, which means you're not being pushed to ditch it after
18 months because something newer and faster has been released.

And because it's not CPU-centric, and because it's not tied to a
specific bus architecture, it means you can upgrade CPUs as you go and
keep that initial investment in the frame.

Case in point - SF15K I'm working on which was bought with UltraSPARC
III CPUs, had some faster ones put in, then some UltraSPARC IVs, then
some faster ones of those, and then some UltraSPARC T2s. All without
downtime, ending up with domains which have different speeds
UltraSPARC IVs in them. This is a box which was bought in 2003 (for
several million) and is still live, in production, serving mission
critical apps.

And, of course, because you control the hardware, and you have a known
set of devices to support, you can not only insanely tune your OS for
that hardware, you can also continue to support that hardware for as
long as it makes sense. The moment you go to generic hardware you lose
that, and that drives your TCO up, and it means you're less likely to
keep the kit in a  critical, live environment for as long.

I'm still plugging Photons into Solaris 10 boxes to get free VxVM
licenses. This stuff is designed to last forever :-)

Cheers,
TOM

- --
Tom Kranz
Email: tom at gaeltd.com    Skype: siliconbunny
Mobile: 07779 149281    Phone/fax: 01344 773240
http://www.gaeltd.com        http://www.linkedin.com/in/tomkranz
-----BEGIN PGP SIGNATURE-----
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkzT4dQACgkQCaTe3ZK74hnBfQCfQbPTLEvVeUyeoBxu1pexIc5M
SwUAnj5gRy400EeOeCAJXmxWZ9VyZkPf
=jhBE
-----END PGP SIGNATURE-----




More information about the OpenIndiana-discuss mailing list