Slashdot Mirror
Oracle To Increase Investment In SPARC and Solaris
An anonymous reader writes "The Slashdot community has recently questioned what Oracle will do with Sun hardware if and when Oracle's acquisition of Sun closes. And it seems that speculation about the future of SPARC hardware has been common among Slashdot commenters for years. That said, it seems newsworthy that Oracle is going out of their way with some aggressive marketing directed at IBM to state clearly their plans to put more money than Sun does now into SPARC and Solaris." MySQL is not mentioned in this ad, perhaps because (as Matt Asay speculates) the EU is looking closely into that aspect of the proposed acquisition.
Larry Ellison would lie to his mother just for fun.
.nosig
I still expect the sale of the ex-sun hardware business to HP to go through, now Oracle have puffed up the price a bit.
"cool kids"?
Wow. You need to get out more often! :-)
"Speaking the Truth in times of universal deceit is a revolutionary act." -- George Orwell
The ad says that Oracle will aim for tight integration with its database. That might be less welcome news for those people who do not use it for Oracle databases.
You need to call him "Hillary Clinton's husband" for the young folks to know who you're talking about.
It's NOT me! It's the meds! I'm on 1000mg of Fukitol.
OK, that fits.
http://www.geoffreylandis.com
Sun is so cash-strapped that investment in Sparc is at low, almost nothing. So it is easy for Oracle to claim they will outspend what Sun does now....all the while looking for a hardware company on which to dump Sparc off. There are plenty of alternatives to UltraSparc based Sun servers, redundancy and SMP can be done more cost effectively
Linux is running servers bigger than Solaris can handle. Linux is running massive databases in corporations. Linux scales to the small PDA all the way to the world's most powerful supercomputers, Solaris can't do that.
Yeah, may be the amount spent on this campaign covered it already ;-)
I'm much more funny, interesting and insightful than the moderators think
Sounds like you haven't read this essay yet.
If they didn't invest in SPARC/Solaris, all their potential customers would run - probably to the very competitors who are likely to buy that part of the business. However, by putting in a small amount of cash, they can appear to be keeping those lines alive, thereby making them worth selling. If they didn't, the brands would die within a year and the money spent on their valuation / acquisition, would have been wasted. So this way, a small amount gambled now could lead to a bigger payback when the business is sold off. Simples.
politicians are like babies' nappies: they should both be changed regularly and for the same reasons
I just heard on the grapevine that Sun is planning in dropping the Netra line of servers (NEBS compliant chassis for telecommunications deployments). Anyone know anything?
Given how little money Sun had, and how many layoffs they were making and had in the works, for Oracle to invest "more" in Solaris/SPARC than Sun did alone wouldn't take much. What would be actually interesting would be information on the updated product roadmap, which is currently a bit sparse and extremely out of date.
SirWired
Not true
While only 1 of the top 500 is running OpenSolaris (and it's using 2.6Ghz Opterons), still, there is nothing inherently unscalable about Solaris or SPARC. I've personally been logged into a 96 core Sparc machine running Solaris 9 and Oracle 10.
My blog
Fixed.
Seems Oracle is willing to spend more on almost any domain except own employees salaries. Not a wise approach if you ask me.
...gis sdrawkcab (usually not responding to ACs; don't bother posting as AC)
No comments about the European Union? THEY are the ones who will decide if Oracle and Sun merge, not us. It'll be interesting to see what happens if the EU says "no"
"I disapprove of what you say, but I will defend to the death your right to say it." - historian Evelyn Beatrice Hall
Oracle is going to need to do a better job with solaris than it did with Unbreakable Linux. If that's any indication what is in store for IBM, then Oracle is just focused on damage control via loud_mouth marketing campaigns. There was a lot of doubt with UL, and now Oracle not only hhas a new OS to manage, but a fairly large collection of high end hardware to peddle. They are not accustomed to so much responsibility. IBM is.
boycott slashdot February 10th - 17th check out: altSlashdot.org
You're saying that Linux is 100% better because it can run on something that is exceedingly rare? Perhaps you might want to try considering some more run-of-the-mill use cases, such as those one run into in any data center and not just Los Alamos's. You know, things like serving, database server, backup server, storage and so on.
Would promising to maintain or increase the investment into MySQL actually smooth things over with the EU?... If I were an Oracle exec, I would strongly encourage support for MySQL as a way to keep people away from PostgreSQL. Articles like this show that PostgreSQL has a lot more potential to win over Oracle customers than MySQL does.
Oracle Needs to Spend $5-billion to make SPARC competitive again in the database workload arena. I just don't see Oracle making those expenditures at a time, when the global economic prospects are still so threatened. Even then it will still take another 3-4 years for a viable design to make it to production. IBM certainly won't sit still, while Oracle is making that spend. I don't think Oracle can rescue Sun from their years of poor spending on Research and Development.
This could the acquisition that starts to make Oracle hemorrhage cash. Look how long it took for Oracle's fusion product of Peoplesoft and Oracle Applications to come to market. Most customers would have left given the choice. Luckily it's much, much easier to change you UNIX platform than it is to change your ERP system.
I want to see a workstation/small server based on the "cool threads" multi-core chips. The servers are nice, but with their rack-mount-only designs, they're either unreasonably expensive or loud (or both). Especially now that VMs are catching on like wildfire, I'd like to be able to throw a ton of RAM and HDD at a single box and have a bunch of zones and VM'd OSs running all at once. Of course, it'd have to have SLI or Crossfire...and allow big graphics to back up that multi-core/multi-threading.
End the FUD
but Linux does all those things - part of my job is replacing Sun servers with Oracle RAC clusters on Linux. Faster, cheaper, just as reliable.
Wait -- Hillary Clinton has a husband? So, this person married to Hillary Clinton, who is she?
My blog
Yeah, but you gotta understand. Without Sun there's just IBM. There's no other vendor in the mainframe business, which is still big business. You don't think the IRS has time or money to manage the size of cluster they would need to operate effectively? So they rely on big iron, which is reliable and redundant and engineered to be that way over 40-50 years of experience. Clusters are garbage compared to a real mainframe. Sure, you have distributed filesystems now, and you can sort of split CPU around, there's management systems, etc, but all of this are ideas that come straight from the mainframe os which does all this "by itself". Google managed to make a pretty cool mainframe from commodity hardware but whatever.
Now, if you're not going to go with IBM for your database, you're probably going to go Oracle. But if you need big iron to run this huge database, you're going to have to go with IBM with z/OS and linux virtual machines or something. Oracle now has viable, proven mainframe line and all they have to do is throw money at it. They'll just move to selling complete packages instead of just DB at the mainframe level. With all this "cloud" bullshit (eg "Mainframe on the internet"), big businesses are interested in managed services and Mainframes have always been vendor managed.
Even IBM minis like AS/400 boxes come with full support from IBM. They monitor the box 24/7. I used to operate them long ago, and I remember that a disk went bad in one of our storage boxes (they had these giant enclosures with over 100 disks in them). Literally the message flashed on my console "SYS01281: DISK ERROR" blah blah blah and I turned around to get the binder to figure out what I had to do. By the time I turned back to my desk my phone was ringing and it was IBM support letting me know a tech would be there within 4 hours to replace the drive. Awesome.
So like, Sun/Oracle can do the same thing, and they can compete if they play their cards right. Oracle has poached a lot of high-end people from IBM in the past so this was only a matter of time.
Regarding MySql: MySql is a toy. Go to where the money is and you will find mainframes still. No one in their right mind would put anything important on MySql. Yeah yeah, facebook pft. If Facebook was making more than a few mil they would switch. Internet hits != money. (I'm talking Fortune 25 money, government money, world organization money, casino money, bank money). So I, for one, welcome Oracle and Sun back to this venue.
Cool! Amazing Toys.
.... Their threaded design provides more threads and cores per Watt than other processors, and designs under development is pushing the further in that direction. And at this point, I am not aware of any Linux distribution that supports Niagara (though there may very well be one).
Databases do not benefit as much by fast single thread execution as they do by very reasonable multi-thread execution. That is because in a database application, or Web application, you want to support many sessions.
And as power and heat become issues in large server farms (mostly running database and web applications), the Niagara line is attractive.... The problem hasn't really been Sun's technology, but Sun's marketing and unfocused management. Larry might be a jerk, but he does know how to focus on making money.
You seem to know what you're talking about, but do you live in an alternate universe where Fujitsu and HP don't exist?
At the DB tier, OK fine, what about the Application tier? There are many enerprise applications that can really benefit from having 8 core chips that can do 8 threads/core. There is some level of efficiency in being able to push 256 threads in a 4U chassis.
If you're talking about single machine SMP, Solaris will go to 256 way SMP on available machines from Sun. Linux can do 1024-way Itanium2. With NUMA architecture things can get even bigger
I really hope this means they are going to pursue Rock aggressively. Knockin' on wood over here...
I would like to see Sun's storage technology emphasized. They have some good products, and good ideas. A little refinement would go a long way. It has been apparent to me for several years that storage is ripe for commoditization, all it is is disks, memory, and fibre channel ports. Whoever can deliver the I/O, reliability, and features without the huge cost of the incumbent enterprise storage vendors should do quite well.
Yay! Finally someone who doesn't just repeat the /. standard: "Sun sucks, use Linux on a giant pile of cheap boxen instead" line.
I salute you, good sir.
SMP can be done more cost effectively
Bullshit. Say what you want about Sun, but noone does SMP more cost-effectively than they.
"The agriculture ministry is not in charge of Gundam" - Japanese ministry official.
but Linux does all those things - part of my job is replacing Sun servers with Oracle RAC clusters on Linux. Faster, cheaper, just as reliable.
So you're just a biased troll. Who else would use "Oracle RAC" and the words "cheaper" and "reliable" in the same sentence?
The only thing reliable about Oracle RAC is the money you spend on consultants trying to keep it running.
Of course, you being one of those consultants means you're quite biased.
This is pretty much spot on.
Oracle did not buy Sun for Java, and they certainly didn't buy it because Sun is profitable. Oracle purchased Sun because Oracles business is Database Solutions, and Sun just happens to have hardware and software IP that can make Oracles position better in that market.
Its really that simple. Oracle is not going to be throwing away Solaris, SPARC, or MySQL, because these are the very things that Oracle purchased Sun for.
"His name was James Damore."
EU should have even bigger shitfits if IBM were to buy them, so if the Oracle sale doesn't go through, I think a VC/Capital Management group would buy them hoping to make a profit by splitting them up.
Do you even lift?
These aren't the 'roids you're looking for.
the application tiers I deal are J2EE servers and other middleware and web servers, so at least for those cases no real advantage to one big machine.
x86 is fast catching up, the six core by 8 processors are out now and soon 8x8 also with hyperthreading, that's going to eat much of the lunch of the traditional unix big-iron realm, as most partitionable machines are carved into that space or below.
HP's Itanic... whoops!... Itanium boxes are in the same league as Sun's SPARC boxes and IBM's POWER products, so without Sun, IBM would not exactly be standing unchallenged. (That said, the PA-RISC to Itanic transition in HP admittedly did not go well...)
In addition, I would go so far as to say that Sun wasn't in the mainframe business either. They made really big UNIX boxes, but did not make mainframes. About the only other mainframe company that comes to mind is the Tandem (now HP) NonStop line of products. Unisys claims to make some, and there are a couple of other tiny players out there. But yeah, IBM pretty much had a mainframe monopoly before, and the still have one now.
Ohh, that post wasn't a long pile of dripping sarcasm?
sure sounded like it. Paraphrased post:
"Mainframes are cool and old and big and fun, clusters are junk even though everyone is using them successfully and scaled out 100x farther than mainframe could ever get"
Paying taxes to buy civilization is like paying a hooker to buy love.
but I've yet to hear MySQL's customer base, which skews toward the technology-savvy Web crowd, fretting about Oracle's impact on MySQL's business.
Could this be a non-issue due to that they can just fall back on Monty's MariaDB? ( community developed, stable, and always Free branch of MySQL )
Reply to That ||
Can you say Fujitsu.. ..
If you're talking about single machine SMP, Solaris will go to 256 way SMP on available machines from Sun. Linux can do 1024-way Itanium2. With NUMA architecture things can get even bigger
I guess I shouldn't be surprised by your ignorance...this IS Slashdot. Let me give you a little nugget of incite:
The number of cores a vendor sells in no way reflects the scalability of the OS that runs on it. By your logic, if Sun only sold 4 way servers, then Solaris could not scale past 4 CPUs.
--AC
probably nuggets of insight are better than incite.
256 cores 512 threads is the last limit I saw published by Sun. Please let me know of any bigger claimed value.
in the real world, the biggest machine that can be bought does put a limit on scalability for any business application. I don't see Sun machines leading in real world benchmarks of common business apps either.
As a Linux-on-desktop user, I am dependent on it. It is a critical ap for me.
OpenOffice could finally break the hegemony of MS Office, if it's not screwed up. I know a few people who are now using it on Windows, by choice, not necessity. But if it's screwed up, it's over.
I hope Ellison sees this as his chance to really stick it to Microsoft. I hope he retains and rewards the existing development team, and starts cleaning and optimizing the existing code base, and if needed dedicates additional manpower and resources. I hope Oracle's capable of doing this without screwing it up.
Heavy R&D spending, plus double the number of sales and support engineers is a lot of additional spending unless they can seriously eat into IBM and/or HP's UNIX business, I'll believe it when I see it.
At the DB tier, OK fine, what about the Application tier? There are many enerprise applications that can really benefit from having 8 core chips that can do 8 threads/core. There is some level of efficiency in being able to push 256 threads in a 4U chassis.
We've recently made the decision to purchase those boxes for use as weblogic servers. The performance sucks. We have had to allocate almost twice the planned hardware to handle our existing loads; and are actually looking to move right back off of that platform with the next tech refresh.
Unfortunately, this is pure anecdote - I don't have any numbers to back this up, I just know the gist of what's been happening since we "upgraded" to SPARC. I can't even tell you the machines that we migrated away from.
you need to do a few search engine queries before making such a silly statement. Even if you wanted to call Sun's big boxes "mainframes", which they aren't, there are over half a dozen big unix-iron companies. And there are several mainframe companies (of which Sun is NOT one)
Even Fujitsu are finding it tough and becoming non-committal:
http://www.channelregister.co.uk/2009/03/30/fts_server_strategy/
They want to gravitate to where the growth is, and it just isn't SPARC.
Linux runs on expensive highly available hardware too. Including real mainframes, which big Sun boxes aren't.
Who's buy SPARC these days? I don't know of anyone, and for similarly priced machines the X86/AMD boxes run circles around SPARC.
And Solaris is completely independent from chip architecture. SPARC Solaris and X86 Solaris are essentially identical, except for the boot architecture. Pretty much the same for the OpenSolaris fork, which is where all the new features are going. (GA, commercial Solaris is essentially a back-port of OpenSolaris, featurewise.)
In one of my old jobs we paid the premium for SPARC only because we had invested pretty heavily in a disaster recovery process based on OpenBoot, and has problems migrating to X86 because of the BIOSes: they were buggy and no two models were the same.
Give a man a fish and you have fed him for today. Teach a man to fish, and he'll say "WHERE'S MY FISH, YOU IDIOT?"
There is this thing called targeted advertising. One makes an ad targeted to a specific demographic. This ad is targeted specifically at companies using Sun hardware and Solaris.
It does not mention MySql because the ad is not targeted at MySQL users. Granted, the set of "Sun hardware and Solaris users" and the set of "MySQL" users can and probably do overlap, but that is beside the point.
Remember, lack of evidence for something (no mention of MySQL in that ad) is not evidence against said thing nor is it evidence for something else (Oracle is planning to do something bad to MySql).
There is no "-1 offended" or "-1 you don't agree with me" mod options for a reason.
Ohh, that post wasn't a long pile of dripping sarcasm?
sure sounded like it. Paraphrased post:
"Mainframes are cool and old and big and fun, clusters are junk even though everyone is using them successfully and scaled out 100x farther than mainframe could ever get"
Clusters can "scale" only because the definition of "scale" is widened to include "lots of little independent things that we can do in parallel".
Sun doesn't really make mainframes and aren't competing with IBM in that market. The problem is that they made their money getting people to spend a lot on server hardware and support for workloads that can be easily run faster and cheaper on x86. 'We are the dot in dot com'? That market disappeared overnight for them as faster and cheaper x86 servers for web applications took hold snd overlapped with expensive SPARC machines.
Yeah. As much as I hate Sun the company, Solaris 9 onwards is actually a near perfect OS - they knew scalability before anyone else did. And it is also the most developer friendly OS - tools like mdb, dtrace are precious and so is the legendary backwards compatibility. The only thing I can crib about is x86 hardware compatibility - that part sucks pretty bad.
probably nuggets of insight are better than incite.
A little off topic, but I wonder if that was meant to be a pun. That's a problem these days. We can't tell the difference between clever and dumb, because we don't really know the person posting.
Solaris has no limit:
/* /*
* max_ncpus keeps the max cpus the system can have. Initially
* it's NCPU, but since most archs scan the devtree for cpus
* fairly early on during boot, the real max can be known before
* ncpus is set (useful for early NCPU based allocations).
*/
int max_ncpus = NCPU;
* platforms that set max_ncpus to maxiumum number of cpus that can be
* dynamically added will set boot_max_ncpus to the number of cpus found
* at device tree scan time during boot.
*/
int boot_max_ncpus = -1;
int boot_ncpus = -1;
Searching for NCPU in the code, you can see that it is set by the CPU driver. Honestly, Solaris has been able to run on multiple architectures for years and the kernel is really not that big. To say that Solaris can't do something without trying it is just idiotic.
We've recently made the decision to purchase those boxes for use as weblogic servers. The performance sucks.
Your team is doing something wrong. Most likely your application suffers from internal bottlenecks as Weblogic flies on a T5120. We went through a lot of the same growth pains. A single thread is for sure slower on a Solaris machine, but overall you are able to do more work because of the massive parallelism.
Not surprising.
There really aren't that many workloads that can take advantage of all the threads on a Sun CoolThreads box. Most of the time, there's only a few threads doing all the work even though a app like Weblogic may run hundreds or even thousands of threads.
And the workloads that can't take advantage of all the available hardware threads get hammered because each individual thread on a CoolThreads box is relatively slow.
That's why Sun has "try and buy".
...yes 'the state it's in": the reference platform for Oracle.
It's always good to have more people at the party. Hardware that
is specialized for heavy workloads is a good an useful thing and
not something that can be abandoned. It's always a bad idea to
only have one option.
A Pirate and a Puritan look the same on a balance sheet.
...and that would be very sound logic.
You sound like someone that's never built applications that have to scale past 4 CPUs.
Until you build it, and show it, it simply isn't so. It's vaporware and wishful thinking.
All of the noise and trying to call others ignorant really doesn't change that.
Whining about NUMA and hundred cpu boxes is especially ironic in a thread about Oracle.
A Pirate and a Puritan look the same on a balance sheet.
Nope. He has something resembling a clue.
Big boxes are EXPENSIVE. There is a reason that companies buy into clustering
and it's not because they like to throw money away. Clustering allows you to
avoid the MASSIVE increase in cost when you go to large scale machines.
You are ALREADY paying a pretty penny for Oracle.
You are also paying by the CPU for the privelege and SPARC always
sucked in terms of performance. Sun was the Microsoft of Unix. Alpha
was remarkably better for those willing to break away from the herd.
x86 is just the new Alpha with less stigma.
Although companies flee to RAC even when they don't bother to move to x86 in the process.
A Pirate and a Puritan look the same on a balance sheet.
"Faster, cheaper, just as reliable."
*
Probably not. ZFS v. ext*, svcadmin v. init scripts, dtrace v. strace (wut?), crossbow v. a non-existent virt. networking schema in linux, solaris streams v. make the network app do the plumbing, etc. etc.
*
Apparently your job is to replace a superior, better-engineered Unix with something piecemeal. For all of our benefit, I hope opensolaris continues to develop such that its rough edges get filed off and we can have a viable alternative to things like RHEL.
-a friendly Linux platform engineer
> Clusters can "scale" only because the definition of "scale" is widened to include "lots of little independent things that we can do in parallel".
That's pretty much what ANY computer does. This includes mainframes.
We're not comparing to "supercomputers" here that need to do some monster calculation that can't be split on a cluster.
Nevermind that we're talking about DATABASES in particular here.
If your application isn't the essence of a bunch of little
independent things that are trivally parallel then you screwed
the pooch.
Ultimately, a mainframe is just an "IO cluster". So the distinct is a bit bogus.
A Pirate and a Puritan look the same on a balance sheet.
And Solaris isn't a desktop OS, so that's not a huge problem. You're far more likely to be running Solaris on a server than on a workstation except for very particular circumstances in the CAD and engineering realm.
My blog
1. opensolaris isn't solaris
2. there can be plenty of other reasons than that one number that might limit Solaris maximum cpu, I've have seen Sun publish the 256/512 number (its on one of the exams I had to take as certified sun systems engineer), though maybe the total has changed. but does anyone make such a machine, I'm not aware of Fujitsu one bigger than that.
2. wrong to say solaris on multiple architectures for years, Sun introduced then dropped Solaris on x86 and ppc multiple times in the past 15 years.
That is simply not true. Sun has long been criticized by analysts for spending too much on R&D. Most of their R&D goes into SPARC development. They spend a lot of money on SPARC because that's where the majority of their revenue comes from. I'm pretty sure it's in the billions. Maybe that is "low, almost nothing" to you.
While its hopeful now that they have stated plans to keep investing time and money, will SPARC and Solaris remain open, or is the plan to close them off? ( if they do, who cares of they invest...)
---- Booth was a patriot ----
Fujitsu's site lists the M9000 as the largest system they sell, 64x quad core with 2 way SMT gives 512 threads so there is no SPARC system bigger than that as it's also the largest SUN sells. The biggest x64 SSI I'm aware of is the ES7000 from Unisys which only goes to 96 cores so Solaris x86 doesn't have a larger system to be tested on either.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
"x86 is fast catching up,"
It doesn't really matter if it's fast. It's still a x86.
The newer P55 systems show promise, but there is more than CPU clock and number of cores to sever performance than you may (or may not) realize.
http://www.dieblinkenlights.com
Anybody running Oracle E-business with more than a few hundred users will, and some just did, for whom I did a capacity plan. We did the sums: a large Sun/Fujitsu was significantly cheaper than a rack full of small boxes, all under-utilized except for the few that were overloaded already.
If you're a small website actioning off things like eBay (;-)) you need the biggest box Sun (or IBM, or in principle H-P) makes to get enough horsepower to do the TP.
Horses for courses
--dave
davecb@spamcop.net
Fujitsu makes SPARCs (really good ones), and H-P is where another poster said "CPUs go to die".
--dave
davecb@spamcop.net
and you're smoking some very powerful shit while doing so. Our 20 linux cluster couldn't keep all nodes running for more than a day or two without a kernel panic, not long enough for any of jobs to complete. Same hardware, running solaris we could keeping all nodes up for as much as two weeks. Dump the SunX86 hardware and went back to "old" Ultrasparc IV+ cpu's and lo and behold uptimes measured in months and execution times improved by 10%. X86 isn't faster or cheaper and linux just plain sucks
This will give
1) Power 7 a run for the money on performance (and will kill every other microarchitecture incl all other x86)
2) Out-RAS SPARC and scale much more flexibly
3) Be available from dozens of vendors including everybody named on this page
4) Run Linux, Solaris, Windows, MacOS, BSD and probably others
5) Be the least expensive 'big iron' architecture available
If I was Oracle, I'd be planning my strategy around this, but I'd certainly not Osborne my SPARC sales by saying so until I'm ready to pull the trigger....And even when I announce, I would say nice things about SPARC to keep those comfortable with it buying until they wise-up. But if you care about price/performance/uptime you'll know where to go.
The only thing I can crib about is x86 hardware compatibility - that part sucks pretty bad.
Well, that and, let's face it, their userland is *incredibly* primitive. Sure, it's POSIX compliant, but that's about all you can say for it. Until you add the GNU toolset, it's deeply painful to use on a day-to-day basis.
If what you say is true, then why did Oracle expend so much effort trying to purchase just the software from Sun (prior to IBM getting involved) while the hardware went to HP?
I generally agree. The writing is and has been on the wall since intel got their act together:
1) intel is winning due to volume and execution
2) there is room for other players in other niches (large=IBM Power, small=ARM, etc.)
The advantage Power7 will still have over intel cpu's is that they are designed for large scale SMP, intel will still not be able to touch Power7 in 32 and 64 proc systems.
actually, I see it is set to 256 unless Makefile of architecture sets it to something else. for the processor used in the big sparc boxes it will indeed by 256.
Is this a troll? Surely you're not serious about RAC
http://www.theregister.co.uk/2009/09/11/sun_sparc_roadmap_revealed/
It's not pretty. Oracle would have a hard time putting less money in, as the only products that look halfway decent are so far out they are pretty much complete vapor.
SirWired
They want to gravitate to where the growth is, and it just isn't SPARC.
Where's the growth at, Intel systems? Pfffft.
If you're talking about single machine SMP, Solaris will go to 256 way SMP on available machines from Sun. Linux can do 1024-way Itanium2. With NUMA architecture things can get even bigger
I'm sorry, are you suggesting there is a 1024-way SMP architecture available, and a 256-way SPARC system is not NUMA?
You're confused, man.
You mean that the company who has been promoting 'Unbreakable Linux', and even sells it's own Red Hat derived Linux distribution, is now starting to increase investments in Solaris ? Arent these two direct competitors ? What's going on here ?
SPARC always sucked in terms of performance.
When it was introduced it had a 4:1 performance advantage over x86. Intel didn't take the lead until the Pentium, and SPARC was in the game until about 1999. It's been downhill since then for single CPU tasks.
There's two sides to that. The classic UNIX userland crap on Solaris is old and crufty - yes!, and GNU userland is much better but there is more than that. Then there are other things Sun has added over the years.. prstat, cfgadm, devfsadm, dladm, fmadm, ptree, pargs, pwdx etc, etc, etc (trying to leave out the big ones everyone already knows about) There is a lot of stuff Linux simply exposes through /proc and /sys which is awful... I can't fathom why there isn't a ptree for Linux, it's such a simple utility and all the long forum posts from Linux users trying to figure out zombie processes is just funny when such a basic utility easily visualizes it.
SAN management in particular on Linux leaves much to be desired. Solaris 10's cfgadm beats any combination of /sys, /proc and dmsetup related misery. One command vs. digging through logs to figure out which LUN and target sdabc is from, which inane devmapper entry wraps it and which subsystem manages it, cating & eching crap into proc somewhere to do the actual SAN level device management.. ugh..
I digress.. GNU utilities can easily be added to Solaris and even replace standard utilities without too much fuss, and SOMEDAY, Linux will have decent system administration utilities.. (not holding breath)
Please let me know of any bigger claimed value.
I know it's pointless to read the thread you're posting in, but the previously mentioned cystorm unveiled at ISU earlier this year has 3200 cores.
The reason you probably don't see Sun boasting about a lot of its HPC stuff is because it doesn't need to. You don't pick up HPC clients like Cedars-Sinai Hospital, UCLA Neuro Imaging, Wolfram Research, and Sandia National Labs by word of mouth, exactly.
This comment is fully compliant with RFC 527.
x86 may be catching up by the numbers, but x86 is still and has always been a terrible architecture. Modern designs (by which I mean arches designed after 1980) deliver similar per-core performance at half or less clock speed. This is especially true for floating point calcs.
I know it's easy to forget sometimes but the 8086 came out in 1978. Everything that's been added since then, protected mode, segmented memory, virtual mode, superscalar, MMX, SSE, hyperthreading, macro ops, and speedstep are all just so many coats of varnish on an ever decomposing turd
This comment is fully compliant with RFC 527.
I don't normally feel the need to post "me too" type of messages, but what you have to say about SAN and Linux storage management hits particularly close to home.
I'm glad someone brought that up. It's a source of pain, that only people trying to use Linux on SAN seem to realize how much of a problem it can be, starting from sd[a-zz] and finishing with dmsetup problems.
They run as a guest, Linux is not the OS running the mainframe.
Then something in your setup is horribly wrong.
SGIs custom kernel for doing that isn't anything like the kernel you have in your desktop system.
The two are as far removed from each other as the solaris and linux kernels are now.
The Linux kernel scales well to ~8 cpus on a single box, clusters and custom software are a different matter.
Read the post, he does not need superior, he needs cheaper version of Oracle RAC.
Comparing System z9 and Sun's M9000. M9000 looks like a joke. Even HP systems aren't close enough. Both Sun's and HP's systems look like glorified Celeron servers, compared to System z.
It's just incredible how much data z9 and z10 can crunch.
How's the shared locking problem working out for you?
Linux can do 1024-way Itanium2
That's 1024 Itanium2 processors, or 4096-way in the usual sense as each processor has 2 cores with 2 hw threads each.
With NUMA architecture things can get even bigger
All big machines are NUMA. And nowadays not-so-big as well. Say a 2-socket Opteron or Nehalem system, that's NUMA too, even though the NUMA factor is obviously much smaller than in those big SGI machines.
Oh, no kidding. Adding SAN to Solaris is by far easier than adding it to Linux, but I never had any problem with either one. The LVM stuff in SLES made it really easy once I got the Qlogic FC driver loaded (which was harder than it sounds because this was on IBM xSeries servers which had some bad BIOS settings related to the PCI bus. Fixing it wasn't hard, but tracking it down took me recalling the old days of having to manually map the I/O address space on PC hardware.
My blog
not relevant, that's a clustered system, in this subthread we're talking of SMP and hardware threading.
I can't fathom why there isn't a ptree for Linux, it's such a simple utility and all the long forum posts from Linux users trying to figure out zombie processes is just funny when such a basic utility easily visualizes it.
Huh, what? The command you're looking for on Linux is 'pstree -pa' (or, 'pstree -paGl' formats it the way I prefer), which produces the same info as 'ptree' on Solaris, but with lines.
My blog