hfss linux
My Ansoft rep says the Linux version can be run under many flavors of Linux. However, I don't want to get into an installation nightmare since I am not a Linux expert. Can anyone recommend a version of Linux that they are using with HFSS v. 11? Or warn me about problems and pitfalls?
Thanks - Jim
Reference on the Red Hat Linux Website is pointing on the HFSS version 8.5 which was released in 2003. So, it is old.
Regarding installation on Linux, I don't know, but ansoft license manager is the biggest issue of all, not the HFSS.
Can;t you use Windows ?
:D
I finally got a response from Ansoft. They support Red Hat Enterprise Linux v3 and v4. v5 just came out recently, but I bought v4.6 to be safe for now. Ansoft has a version that runs under Windows XP x64. They won't answer the question which is the best. I tried to get a recommendation of Intel vs AMD and Windows vs Linux. The beg off answering because "The vendors are also our customers so we can't show favoritism." What is needed is comparison data but with the high cost of a license it is unlikely that anyone but Ansoft can do a real comparison.
I ended up with a dual quad core Xeon system with 8 GB of RAM that I got for a good price. We'll see how it goes. My biggest issue is that I always want to do problems that push the limit on mesh size. I'm hoping version 11 and 8 GB of RAM will make a noticeable improvement in what problems are practical.
Jim
I have HFSS v11 running on Red Hat Enterprise Linux 4. My machine has 8 dual-core AMD Opterons and 64GB of RAM. HFSS crashes almost daily when running problems of any decent size. I asked several application engineers about this, and they said that HFSS is developed first for Windows and second for Linux. So it's much less stable on Linux if you can believe that. Anyways, I recently switched over to Windows Server 2003 Enterprise Edition and have not had any problems. Hope this helps.
Thanks for the reply. I was interested in increasing my computational horsepower. The information I got from Ansoft was very sketchy. They didn't want to recommend one processor over another nor one operating system over another.Their claim was that they didn't want to favor one of their customers over another. How that applies to the operating system I don't know. I'm going to give this a try. I paid enough money that their experts should come to my office and fix the problem.
For my work, it won't do to have a program that can't do eigenmode problems.
How do you like your hardware setup? Is all of that memory useful? I do a lot of big problems with multiple coupled cavities that are limited by memory space and computational time.
Right now I have 8 GB and I'm thinking of expanding to 16 GB. I have room for 64 GB ultimately.
Thanks - Jim
Hmmm. I don't buy their explanation either. The machine is great. I too thought that for as much S as I paid for HFSS that the customer support would be better. In running these huge problems (I often run problems that are 40+ GB RAM), I have noticed a number of issues with HFSS. The app engineers claim that they know about these issues too but it doesn't seem much is being done.
I sure haven't been impressed by the speed of their response to my problem. I think maybe no one has used RHEL 4 with HFSS 11 to calculate eigenmode problems. I'm thinking of taking your advice and switching to Windows. By the way, do you know if there is a limitation on the number of processors in XP +64? Is that why you are using the 2003 Server Enterprise Edition?
Yes, there is a limit on the number of processors. I don't recall what the limit is on XP, but Enterprise will support up to 8 physical processors. By the way, when you are looking for that info, Microsoft only bases the limit on number of physical processors not total number of cores. So you shouldn't have to worry about the quad cores, just whether or not XP supports two physical processors. Do you have the multiprocessor add-on? It's very handy.
I have had the multiprocessor license for HFSS for some time. Is there an XP add-on or were you refering to HFSS? My previous HFSS computer is running with dual Xeons with hyperthreading. Windows Server 2003 Data Center Edition counts this as 4 CPUs, but XP would only count it as two. I am considering switching my HFSS to Windows Server 2003 Enterprise Edition, but it's a bit pricey for me, although the amount of time I've spent dealing with issues in the Linux version is not cheap either.
By the way I am seeing roughly a factor of 3 speed improvement in the processes that use multi processors, using the numbers reported in the HFSS profile.
I have gained a lot with the extra memory because I can run larger problems without spooling to disk. I don't generally let that happen because the problem then takes many hours to solve. It's a little hard to iterate when the solution requires 6 hours. Twenty minutes I can live with.
In any case, it's way better than the finite element 2-D code I used for cavity design on a 12 MHz 286 system with 1 MB of RAM 20+ years ago. But, one thing hasn't changed, the problems I want ot solve are always limited by the hardware.
I was just referring to the HFSS add-on. I think the best plan if you're going with Windows is Enterprise Edition. The way I understand it, Datacenter Edition is priced per processor, and unless you have more than 8 (physical) it's not worth it.
Not sure which machine you're referring to...is the factor of 3 speed-up on your quad-core machine? One thing I thought a bit odd: When I ran on Linux I had a RAID 1 configuration (mirroring) for my two drives. During a run, I noticed the solver gave me a roughly 13-14X speed-up. Not bad for 16 processors. When I switched to Windows I also switched the RAID configuration to RAID 0 (striping) because I wanted to see if that sped up the file writing that occurred during the solve. Now I only get about 10X speed-up. I'm not sure if the solver is less efficient on Windows for some reason or if my thought about the RAID performance was not correct. Anyways, just something else for you to file away for future reference.
When you talk about really large problems in HFSS, how many tetras do that correspond to (direct solver, no iterative)? 100k, 200k?
It's not any problem to do a direct solve on 700-800k tet's. Most of the stuff i do involves many excitations, so the direct solve is faster than the iterative solver even with that many unknowns.
"It's not any problem to do a direct solve on 700-800k tet's" And how much memory does that correspond to? I had 100k tetras with direct solver (second order) and that used 30GB of Ram + 22GB of disk.
My machine has 64GB of RAM. I don't recall the exact RAM usage for those runs, but it didn't use swap space. I think somewhere in the neighborhood of 4-5 million unknowns is nearing my upper limit. The 700-800k tet runs were first order elements.
found this searching for something else, incase someone comes across this searching. HFSS will run on any linux distribution, you just have to change a few lines in the launch scripts. All they do is check if you are running the 'supported' distros. If you comment out this check, it works perfect. Tested on debian unstable. If you contact their support, they may be able to help you, I cant upload the modified scripts since they are copywritten i believe.