Hi Guangya, Thanks for your information. does it mean that I could only managed GPU resources through container based service ?
Warm regards, Jef Chiu 邱征輝 TCCD/ICSD/IT Taiwan Semiconductor Manufacturing Company |-------------------------------------> | Guangya Liu | | <[email protected]> | | | | | | | | 2016/08/23 上午 10:49 | | | | | | | | Please respond to | | [email protected] | | | |-------------------------------------> >---------------------------------------------------------------------------------------------------------------------| | | | | | To| | [email protected] | | cc| | [email protected], [email protected] | | Subject| | Re: Does Mesos 1.0 support GPU ? | | | | | | | | | | | >---------------------------------------------------------------------------------------------------------------------| Just FYI, the current GPU support only support Mesos Containerizer but not Docker Containerizer. There is a patch chain for Docker Containerizer GPU support here: https://reviews.apache.org/r/50123/ Thanks, Guangya On Tue, Aug 23, 2016 at 9:41 AM, David Greenberg <[email protected]> wrote: I've run the GPU support with the compiled packages from mesosphere. Here's a little demo repo that will set things up on an NVIDIA node on AWS: https://github.com/dgrnbrg/mesos-gpu-docker Thanks to Kevin Klues for pointing me in the right direction! On Mon, Aug 22, 2016 at 8:17 PM <[email protected]> wrote: Hi , yes, it seems that NV GPU already supported by Mesos 1.0. I am not sure is there any difference between install Mesos 1.0 from RPM and build from source code . Should we build from source code when we need NV GPU support feature in Mesos 1.0 ? Warm regards, Jef Chiu 邱征輝 TCCD/ICSD/IT Taiwan Semiconductor Manufacturing Company |-------------------------------------> | Sam | | <[email protected]> | | | | | | | | 2016/08/23 上午 08:51 | | | | | | | | Please respond to | | [email protected] | | | |-------------------------------------> >-------------------------------------------------------------------------------------------------------------------------------| | | | | | To| | [email protected] | | cc| | [email protected], [email protected] | | Subject| | Re: Does Mesos 1.0 support GPU ? | | | | | | | | | | | >-------------------------------------------------------------------------------------------------------------------------------| Jef, I think so. Please check release note http://finance.yahoo.com/news/apache-software-foundation-announces-apache-100000615.html?from=timeline&isappinstalled=0 Regards, Sam Sent from my iPhone On Aug 23, 2016, at 8:42 AM, [email protected] wrote: Hi , how could I found GPU resource in Mesos 1.0 ? it seems that the Mesos 1.0 already support GPU . But when I follow the https://open.mesosphere.com/getting-started/install/ and install those RPMs in our servers. It seems that it just found CPU/Memory/Disk resource. But GPU can';t be found by Mesos 1.0 master. The following is logs when I startup Mesos 1.0 agent (GPU installed) with command "systemctl start mesos-slave . I can see the it found GPU in agent log . But Mesos master can't display GPU . Log when start up mesos agent by "systemctl start mesos-slave " ..... I0818 08:58:51.993185 103239 slave.cpp:519] Agent resources: gpus (*):[0-1]; cpus(*):48; mem(*):772718; disk(*):145941; ports(*):[31000-32000] I0818 08:58:51.993273 103239 slave.cpp:527] Agent attributes: [ ] ..... Log when start by command : mesos-agent --master=10.101.1.22:5050 --work_dir=/var/lib/mesos --resources=gpus:2 WARNING: Logging before InitGoogleLogging() is written to STDERR I0823 08:30:57.292623 136165 main.cpp:243] Build: 2016-07-27 20:21:03 by centos I0823 08:30:57.292881 136165 main.cpp:244] Version: 1.0.0 I0823 08:30:57.292901 136165 main.cpp:247] Git tag: 1.0.0 I0823 08:30:57.292914 136165 main.cpp:251] Git SHA: c9b70582e9fccab8f6863b0bd3a812b5969a8c24 I0823 08:30:57.303582 136165 systemd.cpp:237] systemd version `219` detected I0823 08:30:57.303639 136165 main.cpp:342] Inializing systemd state I0823 08:30:57.326640 136165 systemd.cpp:325] Started systemd slice `mesos_executors.slice` Failed to create a containerizer: Failed to NvidiaGpuAllocator::create: Failed to nvml::deviceGetHandleByIndex: NVML has not been initialized I am sure that GPU function work normal on that host. BTW, when I using Mesos 0.28 (package build from source code) agent and connect with Mesos 1.0 master. Mesos master could report GPU status but can't dispatch job on those nodes. Please advise what could I do to manage GPU resource in Mesos 1.0 ? Thank you . Warm regards, Jef Chiu 邱征輝 TCCD/ICSD/IT Taiwan Semiconductor Manufacturing Company --------------------------------------------------------------------------- TSMC PROPERTY This email communication (and any attachments) is proprietary information for the sole use of its intended recipient. Any unauthorized review, use or distribution by anyone other than the intended recipient is strictly prohibited. If you are not the intended recipient, please notify the sender by replying to this email, and then delete this email and any copies of it immediately. Thank you. --------------------------------------------------------------------------- --------------------------------------------------------------------------- TSMC PROPERTY This email communication (and any attachments) is proprietary information for the sole use of its intended recipient. Any unauthorized review, use or distribution by anyone other than the intended recipient is strictly prohibited. If you are not the intended recipient, please notify the sender by replying to this email, and then delete this email and any copies of it immediately. Thank you. --------------------------------------------------------------------------- --------------------------------------------------------------------------- TSMC PROPERTY This email communication (and any attachments) is proprietary information for the sole use of its intended recipient. Any unauthorized review, use or distribution by anyone other than the intended recipient is strictly prohibited. If you are not the intended recipient, please notify the sender by replying to this email, and then delete this email and any copies of it immediately. Thank you. ---------------------------------------------------------------------------
