Posted on 06/04/2009 12:07:47 PM PDT by Ernest_at_the_Beach
I spent several years in projects moving Linux from 64 CPUs to thousands of CPUs. I was personally running Linux on 1024 processor systems with terabytes of main memory a couple of years ago. Massive operating system work has been done on Linux, as well as major library work, over the last five or ten years to get it to work exceedingly well on very large systems.
See further details for example in the Forbes 2005 article Linux Rules Supercomputers.
Or see the Wikipedia File:Operating systems used on top 500 supercomputers.svg file for a graph showing which operating systems run on the worlds 500 most powerful supercomputers, over the years. It went from about 90% Unix (which would include Solaris, I suppose) in 2002 to about 70% Linux in 2006.
In what configuration? Sounds like you’re talking about something like the SGI Altix setup at NASA, yes?
Last I looked, SGI’s kernel changes aren’t yet certified for Oracle deployment. Oracle has their own versions of Linux which they recommend for deployment - based on Red Hat Enterprise distro’s from what I last know. And last I looked, the Linux crowd was pitching a fit about the license Sun had on ZFS being incompatible with the Stallmanesque theology of Linux. Now that Oracle owns Sun, perhaps ZFS’ license changes to allow incorporation into Linux, but that’s probably down the road a bit.
Whether or not Oracle has certified on such systems is up to Oracle.
I was just correcting your earlier, substantially inaccurate claim:
I like Linux, the software, just fine.
What I don’t like is listening to all the theological arguments over the GPL. Where I timed out on the issue was when Linus switched the version control system from CVS over to BitKeeper. The crap that the Linux mob gave McVoy was repugnant. So let’s put it this way: Much like some of the BS that comes with using a Mac, it isn’t the OS or hardware that I don’t like. It is the people that it attracts that I cannot tolerate.
From that perspective, I much prefer the BSD licenses. And truth be told, I like the FreeBSD way of dividing up the work and responsibility - it results in more useful work being done.
When I made the remark of supporting 64 CPU’s, I should have made it more clear that was in the context of supporting Oracle, as Solaris has been doing for Oracle. There’s a lot more to running Oracle and a major database than most Linux advocates would like to admit. (there’s a lot more to commercial applications than most of the Free Software mob would like to admit - but theirs is an epistemological aversion to profit, so they’re consistent). The Linux development mob loves to fiddle with stuff in the kernel - and since they’re not being paid, they’re certainly free to spend their time any way they wish.
However, for a large IT environment running a DBMS, there’s plenty more that has to be there than just a slick kernel. The job ain’t done until the data has been put onto disk.
This is why IBM and Fujitsu continue to ship mainframe iron that is still compatible with software written 40 years ago. That’s why a big piece of Sun’s product line is their storage systems.
For Oracle, their main competition in the largest of database deployments is IBM, and at the high end, they’re competing against z/OS and z9 or z10 series mainframe iron with a similar number of CPU’s on a SMP machine, with software that offers more features than Solaris and a TON more features than Linux. Thousands of CPU’s and tons of GFLOPS don’t enter into the situation - scalability and drop-dead reliability with all the features the database uses/needs, from the CPU all the way down to the disk IO, is the issue for them. From that perspective, it is obvious why Oracle bought up Sun. They have to have Solaris, and for the time being, they need Sun’s server hardware. In the future, they could probably find compliant commodity hardware of the capacity they need in some x86 variant, but they’re going to still want many of the features of Solaris, because it is those features that they’ll need to go up against IBM. In the end, their performance will have to scale based on database transaction performance, not floating point benchmarks.
I’ll grant you that Linux works fine for 1024 or 2048 CPU’s in whatever the latest flavor of atrociously unprofitable floating-point crunch box is. Again, my comments were in the context of scalable Oracle performance.
Absolutely!
Hugely different!!!
Your atrocious theological crap is repugnant.
Have a good day.
By the way, a google search for “oracle unbreakable linux” will provide ample evidence that Oracle finds Linux more suitable than you do for DBMS work, and has so found for at least the last three years.
Exactly my point.
I will, thank you. And a profitable one at that.
By the way, my quip about the “atrociously unprofitable” industry wasn’t a slap at Linux.
Supers have been unprofitable for the entire length of my career in computing (since the early 80’s). The truth is, they grab lots of headlines, they push back boundaries in hardware architecture and design... and they have such a narrow, specialized market that they’re black holes for hardware vendors, sucking in their entire cash flow and balance sheet. That was the case with CDC, then Cray, Cray Research, Convex, Connection Machine, n-cube... now SGI.
Worst of all, many of them are built for government contracts - limited production markets at best, on scant margins.
Were I a VC, and some really bright guys came to me with a business plan on how they were going to make it big with a next-gen super, I’d do everything in my power to convince them to drop the idea and pursue something else.
Supercomputers are a great resume’ items, but they’re horrible at putting money into your retirement account, much like the specialized workstations of the mid/late 80’s. I had a chance to work with Symbolics machines (as a front end to supers at the time, no less), and as much as I loved the hardware which was fast at the time, and much more reliable than something like Sun HW/sw, and the Genera environment was slicker than weasel snot on cold polished tile, I could see it was going to be a dead end, financially and going forward in my career.
Eh ... whatever.
I can’t say as it makes much difference to me anymore. I’m retired now, taking it easy, after some 30 profitable years in highend workstations, supercomputers and Linux. ;).
Already knew about it. And if it were something on which they could base the future of the company, would they have just paid a major chunk of their cash to buy Sun? Probably not. Buying Sun comes with problems — I’m sure that if Linux were ready for prime time, or even reasonably close to ready, they would have let IBM take Sun. Oracle probably doesn’t want to be in the server business. Java is nice, but they could have licensed it. OpenOffice and MySQL - feh. Nice bragging rights, but they probably won’t add much to the bottom line. Sun’s documents/filings indicate that there were two other bidders besides Oracle - and one of them almost certainly was IBM, who is now rumored to be sniffing at RedHat.
For the money they just spent on Sun, they could have hired a whole lot more Linux people than they already have slinging code, paying them hefty salaries to make Linux do anything they wanted, right? They didn’t do that. Instead, they buy up Sun and then they reach out and snatch up Virtual Iron a couple weeks later. It is pretty clear that they want to go head to head against IBM’s z/OS and DB2 offerings - and to get there, they’ll need to either pour a lot of effort into Linux or port over features from Solaris, a combination of the two, etc.
Linux isn’t “unsuitable” for DBMS work. It just ain’t about to compete against z/OS at the high end - yet. Even Solaris is a stretch - z/OS has plenty of features that Solaris still has only on the drawing board. Mainframe-class rollouts are the highest margin accounts Oracle has, and Linux simply doesn’t have the features that z/OS does. Solaris has some of the features, so you could say that Solaris is closer to a head:head match against IBM at the high end, but even it still has a ways to go in features.
BTW — you can run Linux on z/VM and on z9 or z10’s, so Linux is available on IBM’s mainframes. It isn’t the preferred deployment OS for DB2, but it is there for those who want to go that route.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.