Not logged in
Log in now
Create an account
Subscribe to LWN
LWN.net Weekly Edition for May 16, 2013
A look at the PyPy 2.0 release
PostgreSQL 9.3 beta: Federated databases and more
LWN.net Weekly Edition for May 9, 2013
(Nearly) full tickless operation in 3.10
Oracle: SELECT * FROM Sun
Posted Apr 21, 2009 3:28 UTC (Tue) by shieldsd (subscriber, #20198)
Posted Apr 21, 2009 6:17 UTC (Tue) by bronson (subscriber, #4806)
Let's say you're Sun, you want a fast server chip, and you willing to give up single-core speed (what everyone else wants) to go massively multicore. You like the differentiation so you burn a ton of engineer time creating a custom chip and all the infrastructure that goes along with it.
Once you're shipping, you'll find that Intel has already smoked you. Happens every time. They just take a rusty, decades-old design, update it, shrink it, and fab it on a process you won't be able to use for years. And now their $400 chip blows the doors off your custom $1600 silicon for all but the most synthetic workloads.
I agree, we'll see servers become far more specialized to host modern workloads more efficiently. But, when it happens, the successful products will be x86 based. Alas.
Posted Apr 21, 2009 9:15 UTC (Tue) by kripkenstein (subscriber, #43281)
A top-to-bottom solution - Oracle hardware, Oracle OS, Oracle middleware, Oracle DB - will have its place. It'll be a small, but lucrative part of the market. Oracle will at the same time not ignore the bigger market, including Linux. That's my guess.
Posted Apr 21, 2009 12:00 UTC (Tue) by marcH (subscriber, #57642)
A few other non-x86 hardware "niches" are doing not too bad: think for instance GPUs or mobile phone processors (more units sold than x86).
Posted Apr 21, 2009 16:45 UTC (Tue) by bronson (subscriber, #4806)
Obviously ARM, Mips and PowerPC are doing wonderfully in engine computers, industrial control, mobile phones, access points, etc. Why? Because generally the most important factor is power consumption or ruggednes, not raw horsepower. It makes perfect sense to use custom silicon in these applications.
Posted Apr 21, 2009 20:51 UTC (Tue) by flewellyn (subscriber, #5047)
Posted Apr 22, 2009 0:15 UTC (Wed) by leoc (subscriber, #39773)
Posted Apr 22, 2009 4:08 UTC (Wed) by ebs (guest, #30411)
Posted Apr 21, 2009 10:17 UTC (Tue) by hppnq (guest, #14462)
Apple learned it with PowerPC, Intel/HP learned it with Itanium, and Sun has learned it multiple times over. We'll see if Larrabee and Cell can buck this trend but they're looking pretty weak so far.
Never leave Big Blue out of the analysis. ;-)
Oracle is not selling a general purpose system, and my guess is that they will not miss this opportunity to shape an OS to the needs of its flagship product. They must be dreaming of selling and servicing the datacenter-in-a-box, and obviously, IBM is the big competitor here. But I don't think Oracle care as much about the box as IBM does: they will not even try to beat IBM with better and faster processors. If they keep Sun's hardware division, it will be for integration purposes, not differentiation.
Posted Apr 23, 2009 19:22 UTC (Thu) by fuhchee (subscriber, #40059)
That may be the calculus of the thing.
Posted Apr 21, 2009 13:02 UTC (Tue) by gouyou (guest, #30290)
If you care about performance, you must go with the high-volume, off the shelf CPU.
That explains the success of a number of FPGA solutions ... A generic CPU is good for generic tasks but using dedicated hardware can really accelerate some workloads.
Posted Apr 21, 2009 15:42 UTC (Tue) by i3839 (guest, #31386)
Posted Apr 23, 2009 13:18 UTC (Thu) by davecb (subscriber, #1574)
Actually Intel is close to a monoculture, and
without AMD, IBM and Sun, would wander off
into dark corners like the the IAPX432 (shudder) and Itanium (unobtanium?),
without a presence in the 64-socket-and-above
space where Oracle sells a lot of product.
Right now, if you want a big box to run something like eBay or PayPal on, you buy Sun SPARC or IBM Power chips.
In the future, I speculate you'll see AMD competing in
the NUMA space with 32- and 64-socket systems,
implementing the heaviest-used subset of the
x86-64instruction set and faulting to emulate
the leftover dreck (;-))
Posted Apr 23, 2009 17:28 UTC (Thu) by bronson (subscriber, #4806)
Not sure Sparc has really motivated Intel much...? Every Sparc I've ever used has been large, hot, and slow for real-world workloads. But maybe I'm just scarred from having to lug around a Tadpole for a year.
Posted Apr 23, 2009 18:17 UTC (Thu) by davecb (subscriber, #1574)
I think it's probably fair to say that SPARC
and Power were merely goads to Intel, showing
them a mid-range and high-end that they
couldn't achieve. AMD is a punch straight to
the eye (;-))
Posted Apr 22, 2009 12:12 UTC (Wed) by gdt (subscriber, #6284)
Why would this be more successful than Procket Networks, where a group left Sun's SPARC team to create a 40Gbps network processor?
Posted Apr 21, 2009 4:12 UTC (Tue) by wilreichert (subscriber, #17680)
Posted Apr 23, 2009 2:36 UTC (Thu) by jordanb (guest, #45668)
The idea of specialized hardware is that it's specialized for very specific tasks. It will be exceedingly good at them but may suck at everything else.
Posted Apr 23, 2009 13:29 UTC (Thu) by davecb (subscriber, #1574)
Copyright © 2013, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds