Linked by Thom Holwerda on Fri 14th Nov 2008 21:38 UTC, submitted by pantheraleo
Oracle and SUN The world hasn't been kind to Sun for quite a while now, but with the economic downturn, things are getting worse. Sun announced today that it will be laying off 18% of its workforce, or about 6000 people. In addition, it was announced that Sun's software chief Rich Green has resigned for reasons that were not stated, although as part of Sun's reorganization and cost cutting efforts, many departments are being merged, and the software division is being restructured and reorganized.
Thread beginning with comment 337500
To view parent comment, click here.
To read all comments associated with this story, please click here.
RE[4]: Sun performances...
by segedunum on Tue 18th Nov 2008 00:38 UTC in reply to "RE[3]: Sun performances..."
segedunum
Member since:
2005-07-06

But for some multi threaded work loads the Niagara IS many times faster than an x86. You dont trust the benchmarks? They are fake?

No I don't trust them. I don't see any seismic waves on the grapevine as to how much faster these machines are, and the case studies are highly, highly suspicious for the reasons I stated previously. The descriptions some of these people give of their operations (running Fedora in production and then spending a ton of money on Sun kit, for example) simply doesn't stack up. Basically, the only people producing these wondrous benchmarks are Sun themselves and a few organisations Sun are using in their case studies, probably with free kit and consultancy attached for taking part.

It's certainly not an Intel versus AMD shift where lots of verification can be had. Even some benchmarks there have been iffy, but the recent picture has become clear.

Without specifics, the general premise is that you run these machines for very parallel operations. The only way you can possibly make use of this is to split your software operations into a lot of lightweight independent threads of execution, otherwise you are back to work done per single task. Either way, most are going to have to heavily optimise your software to get it to do that. Very, very, very few can exploit multiple cores within a single transactional sequence to any great extent. The simple fact is the vast majority of tasks are pretty single threaded.

Or do you think I claim that the Niagara are in general 20 times faster than an x86 cpu? Where do I claim that?

The premise here is that somehow a Niagara machine is worth any given x86 machine several times over in a specific niche (replace X boxes with Y Niagara machines), benchmarks and specifics of which are thin on the ground outside of Santa Clara. Sun can't really afford to have their hardware people running around playing these games.

Either way, you are wrong. For some work loads the Niagaras are way faster, and neither do I claim Niagaras are that fast in general. Learn to read.

I would suggest you learn to read articles you quote. These are 'benchmarks' issued by Sun and not verified by anyone else, devoid of specifics, and their case studies are merely poster children organisations that have suspect operations. Repeating it won't magically make it true or some statement of fact.

When IBM announce an benchmark, it is not meant to be interpreted as a general statement, right?

No it isn't. Power has failed in terms of raw performance up against x86 machines as well, which is why IBM is keen to avoid crossover where they sell it. Why are you mentioning IBM?

If you think otherwise, you think wrong. Again. I suggest you study some higher mathematics. That will sharpen your thinking skills.

Wow. I would suggest Sun stops spending time trying to tell us about [insert incredible technical innovation that will turn things around here], and spends more time working out why they are laying off thousands and are struggling to break even year after year. Obviously the 'bottom line' skills could do with sharpening.

As of now, you are drawing wrong conclusions. SUN has always been very clear that Niagara cpus are for throughput and not single threaded work.

It's a very, very, very, very, very niche benchmark for the vast majority, with massive effort and retooling required and questionable studies, which has been my point throughout. Sun cannot afford expending R and D money on niche markets unless they are very lucrative. Financial performance. That's the point here.

Ive told you many times Ive run ZFS on 32 bit pentium 4 with 1 GB RAM for over a year. Wrong again. We dont reach you, that's obvious.

"I got Solaris and ZFS up and running on my dinky little 32-bit laptop with 512 MB of memory and, OMG, it didn't die!" does not prove a thing I'm afraid. The BSD guys' experience, and those of many others, are that ZFS will naturally tend to grow unbounded and grab memory as you increase its workload. Its extensive features come at a price, and it depends on whether most think the price is worth paying.

Besides, while you and others argue totally off-topic stuff like this, Sun sheds yet more jobs and is still struggling to get to break-even. Show me the money, as they say.

You try to say that you shouldnt even consider what Niagara boxes can do for your particular work load, because of some strange reason?

I'm not saying that at all. If you have thousands of threads of execution, those threads are quite lightweight, they don't do a lot of number crunching and they are pretty independent and parallel then one of these machines might well be ideal.

Alas, I can't see that being anything other than a niche, the workloads of the majority of organisations don't stay uniform, circumstances change and with advances in x86 machines with multiple cores you question the shelf life of a machine in a year or two that cost your company a lot of money. You're going to look pretty incompetent when your company wants to throw a somewhat new workload at your Niagaras and they fall over...........Niagara falls. Even Sun's consultants are careful where they sell them.

Understand this; Niagara boxes compete with IBM AIX Unix and HP-UX unix, not with x86.

All the major benchmarks and case studies produced for Niagara compare it with x86 and Linux based machines, and talk about workloads such as MySQL and web servers because that's where Sun has been losing out. They don't compete with anything that runs AIX or Power in any way, and that's the problem.

If you believe this then you are very confused about what Sun's target market is for Niagara, which is par for the course really.

Maybe it is you who are naive if you think companies change their important infra structure without testing and doing an analysis first?

Well, they apparently run Fedora in production, so I'd call that naive. Free kit, consultancy and freebies also help. :-)

Frankly, you would get fired if you worked at my company. For incompetence.

Talk to the hand sweetheart. You've spent God knows how many paragraphs talking about everything from ZFS to lovely technical comparisons of Power and SPARC to try and avoid talking about the inevitable, as Sun people love to do - and yet you would ignore Rome burning as you struggle to break-even and your company lays off thousands of people. Quite frankly, you won't get to fire me or anyone else ;-).

If you have a solution that are cheaper and 10 times faster and suits your work load perfect, and still you wont even consider it, then you are incompetent. You make the company loose money it could have saved.

That would be lovely, but alas, in case you hadn't noticed Sun have laid off thousands of workers, again, and they are really struggling to break even year after year. The bottom line is that few out there agree with you.

Unless, you work for IBM or some similar company. Then your comments make sense.

I see an awful lot of people, a great many of them Sun employees, spending a great deal of time and effort banging away on their blogs about IBM, what IBM are saying about them and arguing fruitless points about various meaningless technical benchmarks they have come up with that few are listening to in the hope that will turn things around. You would think they would have other things to do ;-).

Edited 2008-11-18 00:42 UTC

Reply Parent Score: 2

RE[5]: Sun performances...
by akrosdbay on Tue 18th Nov 2008 06:56 in reply to "RE[4]: Sun performances..."
akrosdbay Member since:
2008-06-09


Alas, I can't see that being anything other than a niche, the workloads of the majority of organisations don't stay uniform, circumstances change and with advances in x86 machines with multiple cores you question the shelf life of a machine in a year or two that cost your company a lot of money. You're going to look pretty incompetent when your company wants to throw a somewhat new workload at your Niagaras and they fall over...........Niagara falls. Even Sun's consultants are careful where they sell them.


May be you can explain why Sun's CMT business is growing double and triple digits YoY if there really is no demand for it?

Edited 2008-11-18 07:04 UTC

Reply Parent Score: 1

RE[5]: Sun performances...
by Kebabbert on Tue 18th Nov 2008 10:24 in reply to "RE[4]: Sun performances..."
Kebabbert Member since:
2007-07-27

Ok, if you dont trust official benchmarks about Niagara on SAP's and Oracle's web sites, then there is nothing I can do. We are "discussing" why it would be incompetent to not even consider Niagara for a particular work load if it is 10 times faster. But if you think that the benchmarks and testimonies are a lie, then there is nothing to discuss about. I wonder, what kind of environment are you used to? Lies are quite common, I understand?

I think you should seriously consider study some higher mathematics. Then you maybe could draw conclusions correctly. No mathematician would draw such flawed conclusions as you do.




Wrong conclusion 1:
If web sites still functions perfectly, if not better, with far lesser Niagara boxes, then... what does that imply? Is everything about the Niagara performance a lie, or could it be true? Hmmm... Let me think... Hmmm... Nope, I cant figure the answer out. Can someone else than Segedunum help me? What could the correct answer be? If it is a big lie that you can migrate lots of servers to few Niagara, how can the companies still function after an migration? Hmmm... Difficult question, I must say.

What happens IF the benchmarks that are to be good to be true, were true? Would Segedunum be shocked then?




Wrong conclusion 2:
Segedunum has heard that ZFS requires a lot of memory on FreeBSD to even start/work/whatever. Does that imply that ZFS on Solaris requires a lot of memory? Hmmm... Let me see... Hmmm... Nope. I can not figure that out either. Can someone else than Segedunum help me?

Segedunum has also heard from several different people that ZFS on Solaris doesnt require much memory, but them sources are not to be trusted. Can I spot a slight bias somewhere? Maybe not?




Wrong conclusion 3:
The Niagara boxes are worthless because the Niagaras will not be able to handle a different task with different work load.

Hmmm... Let me think. If... SUN makes it very clear that these Niagara excels at some tasks, but not on other tasks... Does that mean that I can throw what ever task I want at the Niagara boxes? It does mean that, doesnt it? Or? Now I am unsure. This was too difficult for me also.

If I have a knife, that is made for one purpose: cutting, should I expect the knife to handle other tasks as well? Or should I just see the knife as one of many tools? One tool for cutting, one tool for hammering, etc? Nah, I dont know. This was too difficult for me. Or, maybe if I have specialized tools, they excel at a single particular task and I should have multi tools in my repertoire? One specialized tool is 10 times better than a general tool at a particular task? Does that make the specialized tool worthless? I guess it does? Hmmm... Here we see some deep thinking.




Wrong conclusion 4:
SUN doesnt succeed right now. That must imply that SUN's technology sucks and everything is a big lie, right?

Microsoft with admittedly bad technology succeeds very well. And MS is famous for having an excellent and aggressive sales division. So let me do some deep thinking again. If... MS sells best in the world, then their technology must be best in the world, right? And... as SUN doesnt sell too well, their technology must suck right? Ah, yes! I solved it! Yes, it must be so. Nevermind the battle between VHS and Betamax, where Betamax lost despite being better technology.

And besides, The Niagara boxes market share grow very rapid. Like 50% or more each year. There ARE customers wanting those. I wonder why if they suck so badly. So, wrong again, buddy.





I could go on and on, analyzing your earlier posts. But I have work to do. I am doing valuable work for my company. Segedunum, you are so clever man. I wish I had your brains. Where do you find all your bright ideas? They just pop up? *full of admiration*

I understand SUN employees banging away on IBM. I would do that also. Do you know why? Because IBM are stating things that are not really true. If IBM market division didnt exaggerate things, there would be no banging, I promise you. When IBM goes out and says false things, like their technology is best in the world and so - but an Niagara is many times faster at a fraction of the price, I would be pissed too.

Like, when IBM says that an lesser Mainframe consolidates 232 x86 servers. If you scrutinize that, it turns out that the x86 servers must be at 2-3% utilization and the mainframe is close to 100% utilization! And, you can emulate a mainframe on a laptop with "Hercules". An IBM Mainframe MIPS == 4 MHz x86. An 1000 IBM MIPS cpu is roughly equal to 4000GHz x86. And we read everywhere how fast these mainframes are. You can probable migrate one mainframe to one Niagara box. In fact, I think Ive read it somewhere. If I find the link I will post it here.

And still IBM refuses to publish benchmarks. If their technology were so good, why refuse? SUN publishes everything. SUN is clear with the weak points of Niagara, and doesnt try to hide that. Doesnt try to hide anything, like IBM does. If you have nothing good to come with, then you can try to fool everyone instead. That is the reason SUN people gets pissed off. If IBM would stop that, then there would be no SUN banging. Besides, Niagara is quite fast at number crunching also. 1.4 GHz is way faster than IBM Power6 at 4.7GHz. Funny.
http://www.sun.com/servers/coolthreads/t5120/benchmarks.jsp
In floating benchmarks, the Niagara is not that superior, but still it is fast.


I remember when Microsoft claimed that Windows TCO is lower than Linux TCO. When you scrutinize, it turned out that Linux was run on a very expensive IBM mainframe whereas Windows was run on a PC! If you do that, of course Windows is cheaper than Linux! Are you surprised if someone banged away at MS? But If MS stopped, then there would be no banging, I promise you.

Edited 2008-11-18 10:29 UTC

Reply Parent Score: 2

RE[6]: Sun performances...
by Kebabbert on Tue 18th Nov 2008 12:18 in reply to "RE[5]: Sun performances..."
Kebabbert Member since:
2007-07-27

Or I remember when IBM claimed that Power6 had tremendous bandwidth, 300GB/sec or so. When you scrutinized that claim, it turned out that IBM had added the band width in all components in the chip! The different caches plus other things. Maybe IBM doesnt know that, but a chip doesnt have greater bandwidth than the lowest number.

If there is a bottleneck on 10MB/sec, then the chip will have no greater band width than 10MB/sec. You can not add upp all the bandwidth in the different caches. If IBM stopped all these silly claims, then nobody would have anything to complain on IBM.

Seriously, I dont understand why SUN doesnt sell better. If I were a SUN sales man, I bet I would sell lots. I would turn the company.

Reply Parent Score: 2

RE[5]: Sun performances...
by david87656 on Wed 19th Nov 2008 06:06 in reply to "RE[4]: Sun performances..."
david87656 Member since:
2008-09-03

I was recently at Sun and talked with one of the engineers there who explained some of the reasons Sun hardware is relatively fast. This was not a sales pitch, it was just two engineers talking. I'm a software engineer, so there are alot of hardware details that I don't fully understand, but the gist of his description makes sense.

1. Regardless of the chip architecture, Sun optimizes (synchronizes) the speed of all the components so that they are essentially clicking at the same time. There's very little synchronization loss. If you put components on a board, all purchased from different vendors, the synchronization loss is higher.

2. Niagra (Sparc, for that matter in general) is designed to get work done and not boast a high clock speed. That's why you still see alot of these boxes somehow doing very heavy loads. They were running, if I remember correctly, 200-300 thin clients on a couple of Niagra boxes. Try doing that on a Dell.

I think their model makes good sense, if you goal is performance to get work done. In the end, they are cost-effective, but not cheap.

But, we tend to be a walmart kind of society. If you see something that looks similar on the outside and costs 3-5 times as much you tend to be skeptical. But, if you talk to people who actually have implemented Niagra systems, who have the throughput to justify it, generally won't go back to Intel architectures for the really important stuff.

Reply Parent Score: 1

RE[6]: Sun performances...
by Kebabbert on Wed 19th Nov 2008 13:19 in reply to "RE[5]: Sun performances..."
Kebabbert Member since:
2007-07-27

Another point of keeping the clock frequency low, is that a CPU consumes power proportional to the square of the frequency, if my memory serves me correctly.

If you want to keep power consumption low, the most important factor is to keep the clock frequency low. Due to physical laws. And SUN is doing that.

Genius. Imagine the difficult constraints in a world of high frequency CPUs, when SUN was constructing the new family of CPUs: Fast AND low power consumption. How do you combine that? Fast and low power, at the same time?

Well, due to physical laws, you have to keep the frequency down. Ok, lets do that. How do we get speed, if we need low frequency? Many core with virtually no cache misses (an intel x86 server idles 50% under full load due to cache misses, Niagara almost never idles). Do that and you get... Niagara! The key point is "never idle waiting for cache misses". Everyone has multi cores, but the faster the CPU, the higher the latency with a cache miss. IBM Power6 at 4.7 GHz must have the worst penalty upon cache miss, of all existing CPUs due to their high frequency. That is the reason Power6 are comparatively slow. To crank up the frequency is the wrong way to go, this was apparent long time ago. IBM hasnt realised it yet.

Reply Parent Score: 2