Re: [VirtualGL-Users] System Requirements + troubleshooting
3D Without Boundaries
Brought to you by:
dcommander
|
From: DRC <dco...@us...> - 2007-02-12 18:33:59
|
Try re-running the app with tracing enabled: =20 vglrun +tr ansa10 =20 and send me the output off-list. I also notice that you are running = ansa with the "-dl" option to vglrun. Are you sure that is necessary? -dl should only be used if the application loads libGL using dlopen() rather than directly linking to it. =20 As far as network bandwidth usage, this will also vary depending on the application output, but 2-5 Megabits for each 1280x1024 frame is normal = for VirtualGL's default compression mode (so 20 frames/sec =3D 40 = Megabits/sec.) You can reduce this by reducing the JPEG quality. The default settings (Quality=3D95, no subsampling) are chosen such that the JPEG loss is = small enough so as not to be perceivable by human vision (I am working with = NASA to try and quantify that, in fact.) =20 To measure the frame rate in TurboVNC, use TCBench, which is included = with the VirtualGL package (in /opt/VirtualGL/bin, /opt/SUNWvgl/bin, or c:\program files\VirtualGL Client*\) Start some sort of spin animation which changes the pixels at the center of the TurboVNC window on every frame, then run TCBench and attach it to the TurboVNC window. It will = count the frames for about 30 seconds and then report the average = frames/second for that period of time. [-=3DDRC=3D-] =20 _____ =20 From: vir...@li... [mailto:vir...@li...] On Behalf Of S=F6derlund Christian Sent: Monday, February 12, 2007 6:03 AM To: VirtualGL Users Subject: Re: [VirtualGL-Users] System Requirements + troubleshooting Hi again =20 > What version of Linux are you running, and which VirtualGL RPM did you install? =20 rpm -qa | grep Virtual VirtualGL-2.0-20061218 rpm -qa | grep turb turbovnc-0.3.2-20061220 turbojpeg-1.10-20070115 =20 Linux is RHEL3 update 8 meaning kernel 2.4.21-47 and glibc-2.3.2-95.44 =20 > Is Ansa a commercial application, or is it something that we could = obtain for testing? =20 It's a commercial application http://www.beta-cae.gr/ansa.htm but I can perform some tests with some help from you. The application is compiled = on the same system specification as above. Our applications are CAE = calculation applications for pre and post processing of calculation results in 3D = and I have only tested a few so far.=20 =20 The reason for testing this solution is that we could save a lot of = network load by going from pre/posting locally on the clients (10GB result files sometimes) to doing it on the server instead using VirtualGL and = TurboVNC. *Results are produced centrally in our queue system resources and = storage is also centralized, clients are not on the network backbone which means 100Mbit instead of 1GBit. Last week I stress tested the system with one instance of "vglrun glxgears" in fullscreen 1280x1024 and managed to get 40Mbit of network load to the client. A more normal GL screensaver = glsnake produced 8Mbit of network load. I have not measured FPS on the TurboVNC client, is it possible? =20 > We have some plans to conduct multi-session benchmarks in the coming = weeks and months,=20 > but currently we don't have a good sense of how many users a single = GPU can accommodate. =20 > Such a figure would be very application-specific, anyhow. =20 =20 You are right. It's very application specific. CPU:s and memory can = easily be expanded over time but the graphics card is tougher. I need to get a performance monitor somehow for the GPU to be able to figure out how to scale such a solution to these applications. I will research this = subject further. =20 > But if you collect any data on this that you'd like to share, please = post it here. =20 I sure will if I can manage to produce some kind of results.=20 =20 Best Regards =20 Christian S=F6derlund Linux Client/Server Management Volvo Information Technology Department 1200 SE-461 81 Trollh=E4ttan, Sweden * chr...@vo...=20 _____ =20 The information contained in or attached to this e-mail is confidential = and shall not be disclosed to any third party. It is intended only for the addressees named above. If you have received this e-mail in error, = kindly notify us immediately by telephone or e-mail and delete the message and = any copies of it from your system. E-mail is susceptible to data corruption, interception, unauthorized amendment, tampering and viruses, and we only send and receive e-mails on the basis that we are not liable for any = such actions or any consequences thereof. From: vir...@li... [mailto:vir...@li...] On Behalf Of DRC Sent: den 12 februari 2007 12:24 To: 'VirtualGL Users' Subject: Re: [VirtualGL-Users] System Requirements + troubleshooting What version of Linux are you running, and which VirtualGL RPM did you install? Is Ansa a commercial application, or is it something that we = could obtain for testing? My first guess would be that there is some sort of = ABI incompatibility between the application and the version of VirtualGL = that you are using. This might be a problem if, for instance, the = application was compiled on an earlier version of Linux but you are using a version = of VirtualGL designed for a later version. =20 We don't have any hard data yet on multi-user performance. I would generally recommend one CPU per active user on the system, but even this = may vary depending on whether the user is really "stressing" the system or = just sitting there staring at the application. We have some plans to conduct multi-session benchmarks in the coming weeks and months, but currently = we don't have a good sense of how many users a single GPU can accommodate. Such a figure would be very application-specific, anyhow. But if you collect any data on this that you'd like to share, please post it here. [-=3DDRC=3D-] =20 _____ =20 From: vir...@li... [mailto:vir...@li...] On Behalf Of S=F6derlund Christian Sent: Monday, February 12, 2007 3:04 AM To: vir...@li... Subject: [VirtualGL-Users] System Requirements + troubleshooting Hi=20 I'm testing VirtualGL and TurboVNC to see if it fits our demand on 3D visualizations over the network. So far it looks fast and responsive, = only trouble some applications doesn't work (example Ansa). vglrun -dl ansa1210=20 /lib/libc.so.6: relocation error: /lib/tls/libpthread.so.0: symbol = errno, version GLIBC_PRIVATE not defined in file libc.so.6 with link time = reference /lib/libc.so.6: relocation error: /lib/tls/libpthread.so.0: symbol = errno, version GLIBC_PRIVATE not defined in file libc.so.6 with link time = reference glibc version >=3D 2.2 required=20 Right now I have only tested one session against the server but we are = about to stress the machine some more. When I look at the system requirements = in the manual I can't see for how many users these figures are for? Is it = for one session or several? I'm also worried about how many parallel = processes Nvidia Quadro FX 4500 can handle. Any advice on how to monitor the = stress on the GPU? Test server is running RHEL3 update 8 at the moment. I'm also using VNC Session Manager by Norman Rasmussen to handle the sessions (patched = turbovnc client) <http://vncsessmgr.sourceforge.net/> http://vncsessmgr.sourceforge.net/=20 Test client is running Windows XP.=20 Advise is appreciated.=20 Best Regards=20 Christian S=F6derlund Linux Client/Server Management Volvo Information Technology Department 1200 SE-461 81 Trollh=E4ttan, Sweden=20 _____ =20 The information contained in or attached to this e-mail is confidential = and shall not be disclosed to any third party. It is intended only for the addressees named above. If you have received this e-mail in error, = kindly notify us immediately by telephone or e-mail and delete the message and = any copies of it from your system. E-mail is susceptible to data corruption, interception, unauthorized amendment, tampering and viruses, and we only send and receive e-mails on the basis that we are not liable for any = such actions or any consequences thereof. |