How many sites out there are using Linux?

Alex_Pastuhov

Alex_Pastuhov

Legendary Poster
I'd like to have a rough idea as to what percentage of all JDE installations run on Linux backend (Enterprise/Logic Servers only - not if it's purely a DB Server).

If you have multiple Logic/Batch/Enterprise Servers running different platforms, please select multiple boxes.

This would provide counts for different platforms as well, as a bonus. If you have, as an example, 3 Windows and 1 Linux - please check Windows and Linux. I don't see a way to get proper counts - in the example above this would appear as 1 of each, but it's still interesting.

On the other hand, if you wish to vote for multiple clients with different platforms, please check if you can submit multiple votes, because otherwise your vote can distort the overall picture.

This poll completely disregards your database type. Maybe it's going to be in an another poll ;-)

I am hoping the results of this poll may be of interest to others as well. This poll is limited in time for 1 month only.
 
Its also not really an accurate list. Only linux on Intel hardware is supported by JDE...
 
No, no, the results are just fine. They match my setup EXACTLY!
grin.gif
 
When our new System p ships, I'm going to install Power Linux (probably Red Hat) in an LPAR, then configure System pAVE and have a go with the x86 runtime environment within AIX. It should be a good experiment - I'd like to know if the JDE code (and supporting middleware) built for Linux x86 will run on AIX and POWER architecture (2.1 GHz Power5 processors) without taking either a significant or noticeable performance penalty. Just for kicks.

If it works, it does open up the possibility of consolidating Linux workloads onto the POWER platform.
 
So, this would effectively be x86 emulation on POWER architecture? - this should be really slow, I would have guessed. All kinds of emulation will slow things down, but the CPU emulation is probably the worst kind.

Because clearly, x86 machine code will not natively run on any other hardware.
 
Effectively, yes, but technically I think it is more along the lines of translation of system calls. Traditional emulation is of the entire system; in this case the translation depends on a full install of POWER Linux on the server, which AIX can then use to translate and run the x86 Linux applications.

The question on my mind is whether or not the speed and extremely large cache of the POWER5 processor family can compensate for some of the unavoidable performance penalties associated with the dynamic translation (or emulation) of system calls between platforms. It's worth a test.

Apple used essentially the same translation layer when they migrated from PowerPC to Intel processors for their Macintosh line running OS X. I believe IBM may be using the same technology (from Transitive).

I think the greatest benefit would be for certain "surroundware" apps that sit on the fringe of the application environment, serve specific purposes, but haven't been ported to the POWER architecture. The least amount of benefit would probably be those interactive applications such as the web client. Really, I'd wager anything Java would probably slow down to the point of not being worth the penalty. There's obviously a lot more apps available for Linux x86 on POWER, and for IBM to talk about supporting those apps using System p AVE is quite interesting.
 
Not likely: I think Macs are actually running on Intel now, so the machine code is identical and it's system calls, memory mapping, etc. only.

While PowerPC is an entirely different instruction set, hence this will not be enough. It's like running Atari games on a Wintel PC under an emulator - every machine code will have to be translated. This means that the performance will not just suffer - it will be dead - around 100-300 times slower compared to native. Or at least 100 times slower compared to the much lighter emulation under Mac.

Unless IBM switched to Intel CPU's, rebranding them as POWER ;-)
 
Macs are running Intel processors now. However, during the transition phase from PowerPC to Intel, Apple deployed a technology they called Rosetta; it uses Transitive QuickTransit, the same thing IBM is using for System p AVE.

Users could, for the most part, take their apps with them from the PowerPC platform to the new x86 hardware. Independent tests show that PowerPC compiled apps performed between 40% to 80% native speed on the Intel platform. That is really remarkable (about half as fast) and definitely not 100-300 times slower. There are of course limitations, such as applications which directly access hardware, require nonstandard kernel module access, etc.

The point is, some apps don't need to run 100% native speed in order to be effective. What matters is that the tool is available and it may serve as "good enough" until the app can become a native port.

http://www.geekpatrol.ca/2006/02/rosetta-performance/

http://www.macworld.com/article/48991/2006/01/imaclabtest1.html

http://www.transitive.com/news/news_20070423.htm
 
Back
Top