Fedora, Debian, FreeBSD, OpenBSD, OpenSolaris Benchmarks

[jmartin]

[QUOTE=kraftman;110548]@Orvar

He means "we can make some money on it if we'll continue Sun's propaganda, because we've got slowlaris while our competitors, like IBM, don't". ....

Actually IBM, Dell and HP do have Solaris.

http://www-03.ibm.com/systems/x/solutions/os/solaris/

http://h18000.www1.hp.com/products/servers/software/solaris/index.html

http://www.dell.com/content/topics/global.aspx/sitelets/solutions/management/solaris_solution?c=us&l=en&cs=555

Out of curiosity, where did the hostility towards Solaris originate?

[kraftman]

@Orvar

And what does Oracle CEO Larry Ellisson mean, when he say that Solaris is better than Linux, that Solaris is the best Unix out there?

He means "we can make some money on it if we'll continue Sun's propaganda, because we've got slowlaris while our competitors, like IBM, don't". Btw where he said Solaris is better?

http://www.forbes.com/2009/09/22/oracle-zander-sun-intelligent-technology-ellison.html

[kraftman]

@Kebabbert, Orvar Korvar

I would call this SAP enterprise benchmark a relevant link showing that Solaris performs better than Linux on Enterprise work loads. I had earlier posted many such links, proving the same fact. I can repost them if you wish?

Show me Solaris scaling on at least 512CPU machine not some stupid sap papers.

http://www.internetnews.com/software/article.php/3849046/SAP+Microsoft+Gang+Up+on+Oracle.htm

http://rcpmag.com/blogs/lee-pender/2009/11/microsoft-sap-oracle-in-crosshairs.aspx

[kraftman]

@Kebbabert, Orvar Korvar

Aha! Now I understand what you mean with "different RAM amount"! It took a while to figure it out, because you never explained what you meant. I thought for a long time, that you meant: I had posted different benchmarks, one old benchmark where all SAP machines used X GB RAM, and now I try to fool you by posting another benchmark where all machines use Y GB RAM, from another website. I thought you talked about different benchmarks.

You didn't know what were you talking about.

So, not only did the Linux machine use faster CPUs, but also faster RAM

And this means Linux CPU utilization must be higher in this benchmark? You're really making fool of yourself.

At your first arrival you said Linux doesn't scale good on Big Irons. I showed you many times it does and here's some nice example:

http://blog.itaniumsolutions.org/2009/09/extreme-scale-up-with-sgi%C2%AE-altix%C2%AE-4700-and-intel%C2%AE-itanium%C2%AE-processors/

SGI raised the bar again to recapture undisputed leadership in this benchmark with 9,611,262 Business Operations per Second (BOPS) on Altix 4700 with 512 Itanium 9040 cores,1.6GHz and 18MB cache using Oracle? JRockit, a Java Virtual Machine (JVM). The new SGI record is over 74 percent higher than the previous record.

[lsatenstein]

Operatins System Comparisons

I noted that the tests used GPL products such as encryption, decryption and archiving software.

My question is. "What are we testing? Is the compiler efficiency?"

Is the identical code (allow for different include files) being used for each test?

My view is that a measure to be included is space/memory used for each operating system, and for each test.

True enough, a poor scheduling algorithm will alter results and increase elapsed time. Alignment of code on multiple of word size will also affect the look-ahead capability of the microprocessor, whatever the brand.

My feeling is that each version of linux should use the same kernel version, and likewise for competing bsd versions. Then compare linux versions against each other, and bsd versions against each othr, and using averages for linux and for bsd, produce a simple comparison.

Leslie

[jmartin]

OpenSolaris default memory model

As mentioned in discussions on earlier articles, the default memory model will skew the OpenSolaris results when compared to other OSes.
The best example in this round of testing was C-Ray 1.1.

The commands used to compile this test:

gcc -O3 -ffast-math -c -o c-ray-mt.o c-ray-mt.c
gcc -o c-ray-mt c-ray-mt.o -lm -lpthread

On a 64-bit Linux distro, the above commands will default to 64-bit
which provides a significant benefit for floating point intensive tests
such as C-Ray. On OpenSolaris, the above commands will default to
32-bit which uses the older x87 ISA.

To show the difference, I booted bare metal OpenSolaris dev build 131
and Unbuntu 9.04 64 bit on a Toshiba Tecra M10, Intel Core 2 Duo P8400
(2.26GHz), 4GB. This difference between the T9300 CPU used in the
ThinkPad for this article should just be a slightly better clock
(2.5GHz versus 2.26GHz) and double the L2 cache (6MB versus 3MB).

The default gcc version on OpenSolaris is 3.4.3 (admittedly rather
old), Ubuntu 9.0.4 is 4.3.3. On OpenSolaris, you also have the option
to install gcc version 4.3.2. For OpenSolaris, I tested with both
gcc versions. On each platform the tests were compiled without
changing any of the above option except to explicitly define the
memory model by adding -m32 or -m64. The results (in seconds
where lower is better):

.........Ubuntu 9.04 64-bit..OpenSolaris...OpenSolaris
.........gcc 4.3.3...........gcc 3.4.3.....gcc 4.3.2
.........------------------..-----------...-----------
m32......460.80..............467.11........454.22
m64......197.88..............275.35........199.36

When care is taken to compare the same memory model and
relatively similar compiler versions, the performance is
about the same. It certainly isn't the 4X shown for the
C-Ray v1.1 results in the article. (I'm still uncertain
how the test run for the article produced 868.13 on a
faster CPU and larger cache than my result of 467.11
for the default compiler version and memory model).

Regards,

John Martin

[kebabbert]

Kraftman,
Aha! Now I understand what you mean with "different RAM amount"! It took a while to figure it out, because you never explained what you meant. I thought for a long time, that you meant: I had posted different benchmarks, one old benchmark where all SAP machines used X GB RAM, and now I try to fool you by posting another benchmark where all machines use Y GB RAM, from another website. I thought you talked about different benchmarks.

But now at last, I think I understand what you mean with "different RAM amount". You mean that in these SAP benches I posted, from www.sap.com, the machines are differently configured? And the Linux machine has only 128GB RAM whereas the Solaris machine had 256GB RAM? And therefore you wrote "Different memory amount" and didnt explain further what you meant? Ok, now I get it. Well, I checked it up, and let me answer. (I wish you could be much clearer. What is clear to you, is not clear to another person. Half of the posts are me requesting a clarification from you. It is very hard to follow you, you write so terse. Just few words with no explaining: "you know which post I mean" - no, I dont know what you mean. Please clarify!)



Linux runs on HP machine. It uses 128 GB RAM, and eight 2.8 GHz CPUs and is 7 rack unit.

Solaris runs on SUN machine. It uses 256 GB RAM and eight 2.6 GHz CPUs and is 4 rack unit.

The reason Linux uses less RAM in SAP benches, is because HP can use faster memory sticks, they can use PC2-6400! That is the reason. If HP uses 256GB RAM, then HP must use only slow PC2-5300 (which this Solaris machine use)

When you look at page 15 of the quick specs pdf of the HP DL785G6 speed you will find the following note:
"When only PC2-6400 DIMMs modules are installed with a processor then memory bus speeds for 4 or fewer, 6 or 8 DIMMs per processor will operate at PC2-6400, PC2-5300 and PC2-4200 respectively. All other processor and memory configurations will operate at PC2-5300 with 4 or fewer DIMMs and PC2-4200"

The largest memory flavor with PC2-6400 is "8 GB REG PC2-6400 2 x 4 GB". The HP DL785 has 64 DIMM slots. To keep the memory bus at PC6400, you can just populate 32 of them. 32x4=128GB. HP ran the system at the top configuration that allowed them to use faster DIMMs.



So, not only did the Linux machine use faster CPUs, but also faster RAM. And still the Linux machine is less efficient than Solaris, when scaling to as many as 48 cores. I can promise you that the HP benchmarking team would add another 128GB RAM if that would yield higher SAP benchmarks. It is not like HP does not have memory sticks to spare to run a test, when publicising important SAP benchmarks?

If you wish, you can check SAP offical certified benchmarks 2009035 and 2009030. Do you know what is there? Linux and Solaris machines - with the same amount of RAM. And which one do you think is faster? Make a wild guess!




I would call this SAP enterprise benchmark a relevant link showing that Solaris performs better than Linux on Enterprise work loads. I had earlier posted many such links, proving the same fact. I can repost them if you wish?




Apopas,
Yes, you showed me earlier, that all 11 million LoC Linux uses, is not relevant. I dropped that argument. You succeeded. The question is, What does Linus T actually mean, when he says that Linux is bloated? What does Andrew Morton mean when he say that "the code quality is declining"? What does Dave mean when he is saying that "the kernel is going to pieces"? What does Alan Cox mean, when he say that "the kernel should be fixed"? I mean, several Linux kernel developers are talking about bloat and bugs and declining quality. I wonder what they actually mean? It is some code language, they are actually talking about some secret plan where they rob a bank, right? They dont mean what they say, no one of them, right? I find it strange that you all can read their minds and know what they actually mean. How can you do that? Have you Extra Sensory Perception? And what does Oracle CEO Larry Ellisson mean, when he say that Solaris is better than Linux, that Solaris is the best Unix out there? I bet he is talking about what he will have for lunch. Or is it about his football, can you interpret? And what does Barak Obama mean, when he says he wants to send more troops? Is he talking about his backyard? Can you interpret?

[Zetbo]

Where I advertise it? What I do is trying to figure out why there's sometimes something wrong with it, or some benchmark etc.

New thread please.

Exactly what this means. Here: http://www.phoronix.com/forums/showp...3&postcount=21 few posts earlier. Another problems with memory?

Nope, because afaik you never gave me a single relevant link. Why should I bother? Like I said, start a new thread.

Linux system also uses different database.

[Apopas]

No, you did not discuss this before.

When I say before I mean during the time I made that post my link shows...

But here I am quoting Linus T who says the kernel is bloated. Bloat is not the same thing as number of LoC. Is it not a fact that Linus T said the kernel is bloated? Is it not a fact that several Linux kernel developers said that Linux is buggy? Have I not posted such links? Am I lying?

Bloated and buggy are different things.
Back then, you had claimed that Linux kernel is bloated by providing two factss:
a) Linus says so
b) Linux has 11 million of LoC

What's the purpose of providing the LoC if not for justifying bloatness after all?

[kraftman]

You usually advertises Linux. It is ok if you advertise Linux, but it is a bad thing if I advertise SUN? Have I understood you correctly?

Where I advertise it? What I do is trying to figure out why there's sometimes something wrong with it, or some benchmark etc.

Oh yes? In that case, you can quote some of the links that shows that Solaris scales bad, is unstable and buggy. Go ahead. Do it. Or are you just lying, again?

New thread please.

What are you talking about, "different memory amount"? Could you be clearer? When have I posted to different SAP benchmarks with different RAM amount? Could you link to both, so we can see what you are talking about?

Exactly what this means. Here: http://www.phoronix.com/forums/showp...3&postcount=21 few posts earlier. Another problems with memory?

Please give me that link again. In this thread you have not mentioned that about Solaris, but you have claimed that numerous times earlier, without providing any relevant links. Never. So please give that link now.

Nope, because afaik you never gave me a single relevant link. Why should I bother? Like I said, start a new thread.