Hi all, Firstly, thanks very much for the discussion. Since I am a newbie of Mesos, I still can't figure out the answers of my questions.
(1) What is the meaning of "gpu isolation" in Mesos? (2) Per my understanding, the Mesos code can't support GPU directly, like CPU.I should combine with other frameworks, such as Marathon, right? Thanks very much in advance! Best Regards Nan Xiao On Tue, Jan 5, 2016 at 3:52 AM, Anshuman Goswami <[email protected]> wrote: > I have been working on GPU QoS by modifying the GPU kernels. Something along > the lines of this. The basic mechanism is to time slice a grid by running a > subset of thread blocks in one invocation. The main challenge is in > controlling the subset. The modification cost is paid only once during the > first invocation of the kernel. > > Would this be usable? > > On Mon, Jan 4, 2016 at 10:04 AM, haosdent <[email protected]> wrote: >> >> As I know, I could not see cgroup have any plan to add QoS on GPU. >> Currently the general isolation way for gpus is to let different containers >> use different gpu device nodes. >> >> On Tue, Jan 5, 2016 at 1:57 AM, Anshuman Goswami >> <[email protected]> wrote: >>> >>> >>> >>> On Mon, Jan 4, 2016 at 9:50 AM, haosdent <[email protected]> wrote: >>>> >>>> >If two or more containers are sharing the same physical GPU, how does >>>> > the time slicing work? >>>> I thinks this is as same as multiple processes sharing same physical >>>> GPU. >>> >>> Hm...is there any plan to support QoS on GPU? >>>> >>>> >>>> On Tue, Jan 5, 2016 at 1:48 AM, Anshuman Goswami >>>> <[email protected]> wrote: >>>>> >>>>> Sorry as I am not very familiar with docker terminology. >>>>> If two or more containers are sharing the same physical GPU, how does >>>>> the time slicing work? >>>>> >>>>> On Mon, Jan 4, 2016 at 9:17 AM, haosdent <[email protected]> wrote: >>>>>> >>>>>> As I know, GPUs have different nodes, nvidia-docker image isolate >>>>>> different containers by specifying the using node through GPU= >>>>>> environment >>>>>> variable. >>>>>> >>>>>> On Tue, Jan 5, 2016 at 1:08 AM, Anshuman Goswami >>>>>> <[email protected]> wrote: >>>>>>> >>>>>>> I am interested to know the granularity of isolation. Is it at kernel >>>>>>> boundaries? >>>>>>> >>>>>>> On Mon, Jan 4, 2016 at 8:46 AM, tommy xiao <[email protected]> wrote: >>>>>>>> >>>>>>>> currently, i use Docker to support GPU >>>>>>>> >>>>>>>> 2016-01-04 15:15 GMT+08:00 Nan Xiao <[email protected]>: >>>>>>>>> >>>>>>>>> Hi all, >>>>>>>>> >>>>>>>>> I am investigating GPU feature on Mesos, and want to clarify some >>>>>>>>> doubts: >>>>>>>>> >>>>>>>>> (1) This post >>>>>>>>> (http://stackoverflow.com/questions/27872558/does-apache-mesos-recognize-gpu-cores), >>>>>>>>> refers "Mesos does not yet have support for gpu isolation". How can >>>>>>>>> I >>>>>>>>> understand "gpu isolation" here? >>>>>>>>> >>>>>>>>> (2) A recent post >>>>>>>>> (https://mesosphere.com/blog/2015/11/10/mesos-nvidia-gpus/) said >>>>>>>>> "Mesos now supports GPUs.". >>>>>>>>> But from Mesos release note >>>>>>>>> >>>>>>>>> (https://git-wip-us.apache.org/repos/asf?p=mesos.git;a=blob_plain;f=CHANGELOG;hb=0.26.0), >>>>>>>>> I can't see any logs about GPU. From the v0.26 code, I also can't >>>>>>>>> see >>>>>>>>> any special code for GPU. So what is >>>>>>>>> the status about GPU feature on Mesos? >>>>>>>>> >>>>>>>>> Thanks very much in advance! >>>>>>>>> >>>>>>>>> Best Regards >>>>>>>>> Nan Xiao >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> -- >>>>>>>> Deshi Xiao >>>>>>>> Twitter: xds2000 >>>>>>>> E-mail: xiaods(AT)gmail.com >>>>>>> >>>>>>> >>>>>> >>>>>> >>>>>> >>>>>> -- >>>>>> Best Regards, >>>>>> Haosdent Huang >>>>> >>>>> >>>> >>>> >>>> >>>> -- >>>> Best Regards, >>>> Haosdent Huang >>> >>> >> >> >> >> -- >> Best Regards, >> Haosdent Huang > >
