Survey: The Transition to Intel and its Impact on Scientific Computing

In a few weeks the MacResearch crew will be setting sail for Cupertino to nail your Intel transition grievances to the front door of Apple HQ, Martin Luther style. Well, it will probably be more civil than that, but the fact is our team will have the undivided attention of some top Apple Engineers in the near future, and we would like the take the opportunity to act as a proxy for the community. In particular we would like to hear your questions, comments, and concerns about research and scientific computing issues raised by Apple's transition to Intel-based hardware architectures. If you take the time to complete the survey you can be assured that your issues will be conveyed directly to the appropriate individuals within Apple. No strings attached here my friends, take advantage of Apple's commendable willingness to listen to members of the MacResearch community and pull out your digital soap boxes. Feel free to respond to the pre-defined survey questions, or opine to your heart's content in the general comment area. Go ahead, take the survey

Comments

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.

Exchange, a requirement to conitnue in the world of mac

Oddly enough, the one item that would make it easier for our researchers to do their work is not research related at all.

If we had a true exchaneg client for the mac, one that could handle the full range of exchnage functions (schedules, email, calendars, PSTs, addressbooks, etc...) it would greatly improve workflow.

Currently, if someone wants or needs a full exchange client they must use outlook 2001 in classic mode (never a good thing and soon to disapeer anyway) or else find a PC.

Many among you will tell me that exchange is not necesary, but when it is the primary organizing system for a large government facility, it becomes necesary.

So my one request is a full, working, reliable exchange client that will run natively on the Intel Macs.

Thanks.

Exchange

Ditto. This is the one problem that prevents many users here going completely M$ free.
Apple have threatened to make iCal exchange happy for sometime now but haven't come up with the goods -a major disappointment with 10.4

Exchange - Microsoft's problem in some ways?

In some ways this is Microsoft's problem, isn't it? After all its "just" a matter of a port (yes, I know that's not always simple, hence the double quotes).

Fortran

I'd hope I'm not the only one who would like to see Intel Fortran compiler work natively with XCode.

What about Evolution?

gentle persons,

while this is running for the moment under X11-only, what about giving the free "Evolution" are try: http://www.gnome.org/projects/evolution/ ? Since Version 2 it offers exchange-support; Novell provides a diskimage for MacOSX here -> http://forge.novell.com/modules/xfmod/project/?evolution

cheers,

Markus Bongard

Re: Fortran and XCode

Intel Fortran in XCode will be there.

Steve Lionel
Developer Products Division
Intel Corporation

Classic is still useful...

I have an Intel 8051 corss-assembler and HyperCard utilities that run in Classic.

In our business we deal in eternal truth! In the area of standards you don't change things just to get the latest sexy product. Stability is the key to long term reliability, "don't fix what isn't broken!, etc". In setting national measurement standards, we cannot afford a single mistake in results going out. Changing tools should be an avoidable risk.

gentle persons, at least for

gentle persons,

at least for the 8051 cross-assembler -> have a look here: http://www.pjrc.com/tech/8051/index.html if you have gtk installed, this compiles without problem under MacOSX.

Fortran90 debug support

True debugging support for Fortran90 with Xcode would be extremely useful. All the debuggers I have tried with Fortran90 only partially work; they typically provide only partial or no ability to see module variable values, for instance.

FrameMaker for Intel/OS X

Since Adobe decided not to port FrameMaker to OS X, Macs lack a powerful and reliable way to create large, complex technical documents. LaTex has too steep a learning curve for people who already have too much to do. Quark and InDesign lack sufficient features and force users to klutz with too many details. Word is hopelessly unpredictable for documents longer than 100 pages.

Adobe wouldn't have to do a Cocoa port of FrameMaker. Even a universal Unix/Linux/OS X version with I/O through X-Windows would be enough to make users happy.

Perhaps there needs to be a similar meeting with Adobe to share concerns with them.

Try Lyx

I must admit I find TeX/LaTeX to be worth the effort but for those who don't want to take it up, there is LyX. Lyx is a GUI frontend to LateX/TeX that I have my graduate students use for their thesis. It is excellent for structured documents and handles just about anything we throw at it from short papers to 300-440 pages theses. There is a OSX specific version that integrates reasonably well with other OSX applications. It easily imports pdf figures and has live links to external files that are updated at print time.

More information is available at www.lyx.org

Intel transition is going well

From my user/applications support perspective the Intel transition is going surprisingly well. The couple of apps I have tested - VMD and MacPyMol - work quite well, and a number of other applications are coming along also (CCP4 comes to mind). I foresee that many commercial software vendors will drag their feet, which will leave the door wide open for open source projects to gain more foothold.

What could really help us is if Apple would be willing to give an indication what the eventual PowerMac G5 and XServe replacement will be and when. Maybe under NDA, I don't really care. The high performance Intel machines will be a major milestone in science, just as the G5 was when it propelled the Mac from a nice desktop productivity machine to a credible high performance workstation.

A few thoughts

A few thoughts,

Automator:- Could be really useful, but I'd prefer a "Quartz composer" type interface.
Have a look at http://taverna.sourceforge.net/ or http://kepler-project.org/

I'm often using large datasets and it would be useful to store the the data in a SQL database (rather than read/write to files as I do at present), perhaps easier access to SQLite, including from Applescript Studio.

Sadly I'd have to agree Exchange support would be very useful.

Bring out an intel version of xserve ASAP

Applescript Studio allows you to embed a web page in an interface, but if you click on a link it opens in Safari. An option to open within the existing interface.

Scientific computing survey

Hi there, I run myself such a survey with scientists (and myself) at CERN, EPFL, RAL, Orsay labs, LBNL, UCLA, SLAC, INL, etc. Essentially all my previous life scientific relationships.

Here is a condensed version of all the requests split into 3 categories: Fears, Needs, and Wishes. I have discussed these already in Cupertino at the end of last year.

Cheers

Massimo

MacTel transition survey results:

- [ ] Fears
Firewire to go in MacTel ?
keep supporting it properly, the full FireWire device-to-device protocol
Classic support dropped ? Might still be useful
Keep support for PPC platforms as long as possible.
Address the performance problems due to threading issues in the kernel:
http://www.anandtech.com/mac/showdoc.aspx?i=2520&p=8

- [ ] Needs
AFS support: OpenAFS & Arla
Q:Will Leopard - and subsequent vs - break OpenAFS again?
A: stable KPI from Tiger on, hence should be better scenario

X11: Why not integrate it and make it native?
Better integration: cut&paste still problematic
64-bit X11 library: the X11 protocol allows 64 bit client programs to work with 32 bit servers and vice versa. Xfree86 already support 64 bit apps.

Hardware acceleration

Intel compilers, including FORTRAN
could we have details and gdb support as well
note: Intel Beta programme has been launched and available

GCC support with SSE(1-3) optimization
as much as possible cover gcc linux distro options and provide equivalent or very same

Java 1.5 and keep updated
Optimized libraries BLAS & LaPack
Basic Linear Algebra Subprograms
Linear Algebra PACK

Good intel FORTRAN compiler

- [ ] Wishes / Hopes

AutoCAD for OS X

Full Python support
as well in Xcode
Integrated Virtual Machines and access to Virtualization
support Xen?
no need of VPC
Connection/Discorevy tools for NFS
Carbon Emacs by default, not terminal one. Native X11
Faster MP platform: both desktops and laptops
note: seems we got this covered already
Longer battery life, lighter laptops - at least do not increase weight
Dynamic loader (ld): as Linux-like as possible on lazy linking when dldloading modules

Native support for Virtual desktops

64-bit and better than the Quad.
The system must support Cilk by MIT: http://supertech.lcs.mit.edu/cilk/
Support for both MPI and PVM
Needs for a vector processor for each core and it should be 64-bit based.

Exchange

Has anyone tried
http://www.snerdware.com/groupcal/faq.html

The value of using Apple

The value of using Apple hardware to run OS X vs. using generic Intel/AMD boxes running Red Hat will have to be made very apparent. Currently, some applications take advantage of Altivec, such as BLAST, meaning that the Xserve PPC is of particular value.

However, that advantage will be lost with a move to Intel.

Therefore, what advantage will Apple have?

Cause it's not any of cost/support/documentation/ubiquity. The Server Admin tools are nice and unified, but of little value after initial configuration has been done and a headless cluster is single purpose computational. For that benefit alone, we won't be willing to pay 20-30% more--that would simply be the reason to go with Apple hardware if all things else are equal, and to make up for the complication of using a less common OS.

iLife on the desktops is a compelling reason to go with iMacs, still. However, what is the defining reason to use Apple Server hardware in a lights out, command line, grid environment vs. Linux?

Entourage--good Exchange client

Entourage 2004 on OSX 10.4 (part of Office 2004) connects seamlessly with our Exchange 2003 servers. We get full calendaring (with shared free/busy info), public folders, shared global address book,

It uses Webdav to connect, so does not need a VPN to get through firewall. Entourage doesn't use PST's but can import all their information.

I'm a long-time Windows Outlook user and recent Mac user, and find Entourage to be very capable.

Java issues on Intel

I recently moved to a MacBook Pro from PPC-based computers. I am running several custom research applications written mostly in Python and C. Porting everything over required some, shall we say, "creative endeavors". I now have all my own code working, however, and most of the other utilities I use work as well.

My first observation is that these chips are in fact very fast when running locally compiled code. My MacBook 2.0 outruns my dual G5 2.5 powermac on most of my jobs! I have, however, hit two substantial snags.

The first snag is that Rosetta is painfully slow, and currently Matlab requires it. This machine, which outperforms a G5 on native code, is about equivalent to my g3 700 iBook when running matlab under Rosetta. I have tried installing the intel binaries of Matlab (compiled for linux) but these seem to make Kernel calls tat are not compatible with Mach-O Intel. I can't think of any solution to this but to hope that Mathworks supports the new platform. Personally, I don't use Matlab, or much other closed-source code, so this isn't an issue, but it's a reason why I have to warn many of my scientific coworkers off the Intel Macs. They need Matlab. (BTW: Contrary to some rumors I've heard on the 'Net, I've had no trouble running Microsoft Word, Adobe Acrobat, and Photoshop under Rosetta. I'm sure photoshop would be slow under heavy image processing, but for simple Office type use; cropping photos, writing documents, etc, these apps work fine in emulation.)

The second issue is that Java confuses Rosetta. Ironically enough, the only applications that I can't get running at all on Intel are Java Apps! This includes the Matlab desktop, and the drivers for several of my input devices (most importantly a FingerWorks keyboard). I'm pretty sure I understand what is going on here. Since Java is "platform independent", Rosetta dose not provide a wrapper around Java programs. "100% Pure Java" programs actually are platform independent, and run fine on Intel, but many real Java programs contain system calls, binary extensions, or other "impure" code. These Java apps will not run on Intel, and, because they are Java, Rosetta will not emulate PPC for them so they fail outright.

If I call one of these Java Apps with console output, the first line I get is:

[JavaAppLauncher] application launched with ppc-thin application stub. Using native application stub instead.

Eventually I get an error like this one (exact error depends on the App):

Invocation of this Java Application has caused an InvocationTargetException. This application will now exit. (LAX)

Java issues on Intel

It looks as though Apple is aware of this issue, but they don't offer a solution, only encouragement to developers that they can't mix Java and PPC native code.

http://developer.apple.com/qa/qa2005/qa1295.html