I'm a member of
Quicksearch
Codenews Search
Disclaimer
The individual owning this blog works at Sun Microsystems GmbH in Germany, a subsidiary of Oracle. The opinions expressed here are his own, are not necessarily reviewed in advance by anyone but the individual author, and neither Oracle nor any other party necessarily agrees with them.
Navigation
Categories
About this blog- Apple
- iPhone
- Aviation
- Bahn
- Begriffe des Lebens
- Blogosphere
- Braindump
- Business Travel
- CANBOX
- CeBIT
- CEC
- Computing
- del.icio.us
- Fundsache
- General
- HHOSUG
- i hate sundays
- Kochen
- Mindfuck
- Movies
- Music
- On Benchmarketing
- Persoenliches Pech
- Photographie
- policy of ...
- Privacy
- Recommended Blogs
- Security
- Solaris
- Strange Days
- Sun/Oracle
- suncec2007
- suncec2008
- Technology
- The IT Business
- Toolbox
- Travel
- Work
Do more with less
Monday, December 7. 2009
Trackbacks
Trackback specific URI for this entry
No Trackbacks
Comments
Display comments as
(Linear | Threaded)
As usual, only trust the benchmarks you manipulated on your own 
What do you think had a larger impact: OS or twice the amount of memory? The Sun box had 256 GB of memory, whereas the HP boxes only ran with 128 GB.
What a coincidence that even the article on the SAPonSun blog doesn't mention this little fact.
What do you think had a larger impact: OS or twice the amount of memory? The Sun box had 256 GB of memory, whereas the HP boxes only ran with 128 GB.
What a coincidence that even the article on the SAPonSun blog doesn't mention this little fact.
Jan explained me once that memory is just a factor in this benchmark when you don't have enough (as in "trying to run this benchmark with 8 GB" ) but having more memory than the minimum doesn't help.
It runs horrible without enough memory, it runs optimal with a certain size of memory and it don't get faster with more.
) but having more memory than the minimum doesn't help. It runs horrible without enough memory, it runs optimal with a certain size of memory and it don't get faster with more.
I see, thanks Joerg.
So assuming that Sun chose a reasonable sized machine and that the benchmark improved considerably by using 256 GB of memory, one could also state that Solaris needs twice the memory to be just 25% faster than a competitor
Without the comparison to a identically configured system with just 128 GB of memory, it's difficult to judge. Really a shame that there are no benchmarks of otherwise similar boxes that only differ by the OS.
But the CPU utilization of the HP system shows that Windows has a hard time to keep all the cores under pressure. Solaris is close to 100%, Windows Server 2008 only at about 87%.
And then there's this recent Linux kernel patch that improved ffmpeg performance on a quad-core by ~30%...
So assuming that Sun chose a reasonable sized machine and that the benchmark improved considerably by using 256 GB of memory, one could also state that Solaris needs twice the memory to be just 25% faster than a competitor
Without the comparison to a identically configured system with just 128 GB of memory, it's difficult to judge. Really a shame that there are no benchmarks of otherwise similar boxes that only differ by the OS.
But the CPU utilization of the HP system shows that Windows has a hard time to keep all the cores under pressure. Solaris is close to 100%, Windows Server 2008 only at about 87%.
And then there's this recent Linux kernel patch that improved ffmpeg performance on a quad-core by ~30%...
Good eye, that benchmark almost looked identical.
This is funny. Technically, Sun is trying to push for X4640 server vs the HP DL785G6. If that's the case, HP should run a benchmark using Solaris too. I do believe that Solaris has higher vertical scalability(roughly 10% advantage) at this time than Linux does at 32cores and beyond. So in terms of hardware, DL785G6 and X4640 are interchangeable, and unfortunately both platforms will be obsolete in 3 months when Nehalem-EX and Opteron 6100 hit the market.
JoChen, one thing you will learn is that Sun people really don't like people like us who look at the hard cold numbers. So take it easy before Joerg calls you a HP schill or whatever schill Sun doesn't like.
This is funny. Technically, Sun is trying to push for X4640 server vs the HP DL785G6. If that's the case, HP should run a benchmark using Solaris too. I do believe that Solaris has higher vertical scalability(roughly 10% advantage) at this time than Linux does at 32cores and beyond. So in terms of hardware, DL785G6 and X4640 are interchangeable, and unfortunately both platforms will be obsolete in 3 months when Nehalem-EX and Opteron 6100 hit the market.
JoChen, one thing you will learn is that Sun people really don't like people like us who look at the hard cold numbers. So take it easy before Joerg calls you a HP schill or whatever schill Sun doesn't like.
Oh ... just in case i didn't told it to you: I don't think that you are an Intel shill anymore ... but i'm not sure that the moniker i've put at your name has a better reputation
Thank you. Apology accepted. The word "schill" is a retarded word. It is usually used by a person who lost the argument who feel that they still want to give their opponent one last poke to save face.
Look Joerg, both you are I love ZFS and Solaris. Sometimes it is better to have a critical friend than an ass-kisser enemy.
Anyways, does Sun have balls to sell Fishworks as a software package like Apple sells XSan? At least let people have an X4170 based Open Storage server please? CC this note to Shapiro. I think he understands. Think about it this way, would you like to print $100 bills 100 times, or $10 bills 100K times.
Look Joerg, both you are I love ZFS and Solaris. Sometimes it is better to have a critical friend than an ass-kisser enemy.
Anyways, does Sun have balls to sell Fishworks as a software package like Apple sells XSan? At least let people have an X4170 based Open Storage server please? CC this note to Shapiro. I think he understands. Think about it this way, would you like to print $100 bills 100 times, or $10 bills 100K times.
facepalm
Hopeless.
You've got the facepalm for two things:
1. My comment wasn't meant as an apology at all.
2. Do you really think your idea of an "unbundled" Fishworks software is entirely new and never discussed inside of Sun? That there weren't customers that asked for the software without the hardware? And that there weren't any studies to find about the economic prospects of such a move? Like market analysis, calculation of additional costs to develop, productize, support and sell such a product. As i tried to explain to you several times: There are costs in a product not directly visible.As long as you don't integrate this in your thoughts, any further discussion is futile.
There is already somewhat like an unbundled Fishworks ... it's called OpenSolaris.
1. My comment wasn't meant as an apology at all.
2. Do you really think your idea of an "unbundled" Fishworks software is entirely new and never discussed inside of Sun? That there weren't customers that asked for the software without the hardware? And that there weren't any studies to find about the economic prospects of such a move? Like market analysis, calculation of additional costs to develop, productize, support and sell such a product. As i tried to explain to you several times: There are costs in a product not directly visible.As long as you don't integrate this in your thoughts, any further discussion is futile.
There is already somewhat like an unbundled Fishworks ... it's called OpenSolaris.
1. It is just not nice to call people a schill. PERIOD.
2. I do realize that OpenSolaris is the unbundled software. I am using it. I love it. Fishworks really is just a DTrace GUI on top of OpenSolaris. The point is, I am not opposed to paying a $10K license for Fishworks. (The $10K is derived from the price differential between the X4140 and 7310 and X4540 and 7210). The point is, after I paid a $10K license for Fishworks, I don't like the fact that the rest of the hardware choices are being taken away from me(ie. having to use double/triple stacked premiums for Hitachi UltraStars, and Stec SSDs, where better alternatives exist). So the open storage line of business isn't that "Open" after all.
Coming from #2, I think Sun's open storage is a game changer, but not in its current form. It needs to be more Open in hardware choice. And releasing the Fishworks as a software package on top of ANY SunFire server would be a great step toward that goal. ZFS also needs L2ARC persistence, L2ARC lzjb compression, ondisk and L2ARC encryption, BP rewriter before it can be used in any serious manner in the enterprise world.
I will just wait a little longer. All those things will be eventuality.
2. I do realize that OpenSolaris is the unbundled software. I am using it. I love it. Fishworks really is just a DTrace GUI on top of OpenSolaris. The point is, I am not opposed to paying a $10K license for Fishworks. (The $10K is derived from the price differential between the X4140 and 7310 and X4540 and 7210). The point is, after I paid a $10K license for Fishworks, I don't like the fact that the rest of the hardware choices are being taken away from me(ie. having to use double/triple stacked premiums for Hitachi UltraStars, and Stec SSDs, where better alternatives exist). So the open storage line of business isn't that "Open" after all.
Coming from #2, I think Sun's open storage is a game changer, but not in its current form. It needs to be more Open in hardware choice. And releasing the Fishworks as a software package on top of ANY SunFire server would be a great step toward that goal. ZFS also needs L2ARC persistence, L2ARC lzjb compression, ondisk and L2ARC encryption, BP rewriter before it can be used in any serious manner in the enterprise world.
I will just wait a little longer. All those things will be eventuality.
Hmm ... i know this is problem. IBM try to use this point that we don't downsize the system to the max at more than one situation. But most people in the business are aware of the performance characteristics of the SAP benchmark in regard of memory, so it doesn't hurt where it matters most.
I should explain that those ultra new systems are shared between group performing benchmarks. The next benchmark on this system may think different about the memory size and suck every single byte in the system. Thus i understand why they used a 256 GB system.
But i don't think that we will see really identical benchmarks. HP won't use Solaris on their systems as they don't have interest in pushing solaris and Sun wants to show that SAP runs best on Sun, and I hope you understand why Sun don't want to invest into the Linux benchmark when we have our own operating system.
This is the reason, why otherwise than TS wants to suggest, comparing numbers in benchmarks are not just about numbers, it's about experiences with benchmarks to know the influence of a number to the result.
PS: I'm aware of several performance improvements for ffmpeg ... could you provide a link to the one you actual think about?
I should explain that those ultra new systems are shared between group performing benchmarks. The next benchmark on this system may think different about the memory size and suck every single byte in the system. Thus i understand why they used a 256 GB system.
But i don't think that we will see really identical benchmarks. HP won't use Solaris on their systems as they don't have interest in pushing solaris and Sun wants to show that SAP runs best on Sun, and I hope you understand why Sun don't want to invest into the Linux benchmark when we have our own operating system.
This is the reason, why otherwise than TS wants to suggest, comparing numbers in benchmarks are not just about numbers, it's about experiences with benchmarks to know the influence of a number to the result.
PS: I'm aware of several performance improvements for ffmpeg ... could you provide a link to the one you actual think about?
If memory would be the limiting factor, HP would have used more. 
OK! I must admit, i have looked at the hp-page and the machine is limited to 128GB. Funny thing, the SUN machine isn't.
OK! I must admit, i have looked at the hp-page and the machine is limited to 128GB. Funny thing, the SUN machine isn't.
That's not true. The HP ProLiant DL785 G6 can use up to 512 GB RAM. Also the older G5 model can use up to 512 GB RAM.
You are correct.
PS: A 8 socket machine maxed out at 128GB nowadays would be a really funny construction. Would be useless for many tasks like virtualisation for example.
PS: A 8 socket machine maxed out at 128GB nowadays would be a really funny construction. Would be useless for many tasks like virtualisation for example.
Indeed. But I never would use a DL785 for virtualization. I would recommend blades, 1HE or 2HE servers. Better use more servers with fewer VMs, then fewer servers with more machine. Scale out then scale up.
This is question of personal taste how many hypervisors you want to manage. Some people prefer few larger systems, others want many small systems. I have observed both in customer situations.
OK! I have looked at another machine.
But that supports my first argument: HP did not use more memory, since the machine will not perform better with more memory.
But that supports my first argument: HP did not use more memory, since the machine will not perform better with more memory.
Maybe... But that's speculation.
I wouldn't call this speculation.
Absolutely nobody would harm benchmarking results if he or she could yield better ones just by adding memory and there aren't other constraints (like price/perf metrics).
That would be like running a race and stopping 10 meters in front of the finish.
Absolutely nobody would harm benchmarking results if he or she could yield better ones just by adding memory and there aren't other constraints (like price/perf metrics).
That would be like running a race and stopping 10 meters in front of the finish.
I think we run into scaling issues here. Clearly Linux still has a bunch of issues scaling to more than 16 Cores plus (at least last time I dealt with it) MaxDB had scaling issues above 8 cores.
In Windows 2008R2 scaling boundaries are now pretty much gone on such systems but that still doesn't make SQL Server Scale better.
In Windows 2008R2 scaling boundaries are now pretty much gone on such systems but that still doesn't make SQL Server Scale better.
From the comment chain:
"The X4640 can use HT3 with all it's features, while the HP box is still using HT2. So, there is a slighly impact of the memory, but not it's amount (it's not completely used) but it's interconnect to all processors."
"The X4640 can use HT3 with all it's features, while the HP box is still using HT2. So, there is a slighly impact of the memory, but not it's amount (it's not completely used) but it's interconnect to all processors."
Links in this article
The LKSF book
The book with the consolidated Less known Solaris Tutorials is available for download here
Twitterfeeds
twitter.com/c0t0d0s0
just blogged: links for 2010-03-19: Gedanken eines Fliegenden: Freikoerperkultur ... http://bit.ly/c21ARU
twitter.com/codenews
6935782 need to manually increment build number one more time http://bit.ly/aMqEbX
twitter.com/SunPatches
128365-04 - Sun Crypto Accelerator 6000 1.1: Driver Patch. Available for SPARC since Mar/19/10. http://bit.ly/agl9Nw
twitter.com/SolPatchesX86
118192-04 - SunOS 5.9_x86: gtar patch. Available since Mar/19/10. http://bit.ly/cbnoJ7
twitter.com/SolPatchesSPARC
118191-04 - SunOS 5.9: gtar patch. Available since Mar/19/10. http://bit.ly/cb2Drj
just blogged: links for 2010-03-19: Gedanken eines Fliegenden: Freikoerperkultur ... http://bit.ly/c21ARU
twitter.com/codenews
6935782 need to manually increment build number one more time http://bit.ly/aMqEbX
twitter.com/SunPatches
128365-04 - Sun Crypto Accelerator 6000 1.1: Driver Patch. Available for SPARC since Mar/19/10. http://bit.ly/agl9Nw
twitter.com/SolPatchesX86
118192-04 - SunOS 5.9_x86: gtar patch. Available since Mar/19/10. http://bit.ly/cbnoJ7
twitter.com/SolPatchesSPARC
118191-04 - SunOS 5.9: gtar patch. Available since Mar/19/10. http://bit.ly/cb2Drj
Web 2.0
Contact
Networking
open.bc
LinkedIn
Facebook
My photos
- mail: joerg@c0t0d0s0.org
- icq: 144000169
- Me, myself and I
- openBC
- My del.icio.us
- c0t0d0s0@twitter
- c0t0d0s0@last.fm
- wiki@c0t0d0s0
- Amazon Wishlist
Networking
open.bc
My photos
Syndication
Tagged articles
AMD Apple avs Bahn Blogging Blogosphere braindump Business Travel CeBIT cec cec2006 CMT del.icio.us deutsch dtrace fliegen Fundsache General Hamburg IBM i hate sundays Intel iscsi jumpstart Links Linux lksf Mindfuck Movies Music Musik Niagara Opensolaris Opteron Photographie policy of ... Politik Security Solaris storage Sun suncec2007 sunw t1 The IT Business Ultrasparc ultrasparc t1 Wirtschaft Work ZFS
Comments
about Who are you?
Sat, 20.03.2010 02:15
Ich bin im Rahmen der Diskussi
on um das Zugangserschwerungsg
esetz auf dein Blog gestoßen.
Als Linux-Begeisterter d [...]
Sat, 20.03.2010 00:32
The article doesn't explain wh
y the adquisition of Sun is go
ing to be a sucessfull. It onl
y says that we all know: [...]
Fri, 19.03.2010 20:58
Well, I am being paid to take
care of Solaris 10 systems and
my company will continue to u
se it. But the relativel [...]
Fri, 19.03.2010 17:36
Actually I am curious to know
what would have happened if th
ey objected.
Fri, 19.03.2010 17:31
I agree, it has been a very il
l and stupid waiting...full of
stupid statements...I was so
much waiting for them to [...]
Buttons
This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 Germany License
