You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@mxnet.apache.org by Yuan Tang <te...@gmail.com> on 2020/04/02 14:49:55 UTC
Kubeflow support for MXNet deployment
Hi community,
We are currently looking into adding support for Kubeflow deployment of
MXNet via the existing kubeflow/mxnet-operator:
https://github.com/kubeflow/mxnet-operator. If you are interested in
helping out and getting involved, please let us know in this issue:
https://github.com/kubeflow/kubeflow/issues/4805.
Cheers,
Yuan
--
Yuan Tang
https://terrytangyuan.github.io/about/
<https://terrytangyuan.github.io/about/>
Re: Kubeflow support for MXNet deployment
Posted by Yuan Tang <te...@gmail.com>.
I just added an initial list of adopters of MXNet Operator here:
https://github.com/kubeflow/mxnet-operator/pull/63
*Commnity - If your company uses MXNet Operator, please let me know in the
PR*.
On Mon, Apr 6, 2020 at 9:08 AM Yuan Tang <te...@gmail.com> wrote:
> Definitely. Thanks for asking and I am happy to provide more background.
>
> The MXNet Operator has v1 controller. However, there has no formal release
> on GitHub and it has not been included in Kubeflow by default yet. Though
> users are welcome to use it as a standalone k8s operator. This will
> hopefully be something the community will help drive forward as the initial
> authors are not active. I don't have a good idea of who's using it yet
> externally but I'll start by adding an ADOPTERS.md. *If your company uses
> MXNet Operator, please let me know*. I will post the link to the PR once
> it's ready.
>
> Another easy-to-use option on k8s would be using Horovod + MXNet with *Kubeflow
> MPI Operator* [0] (I just added a section in MXNet's doc on this [1]),
> which already has wide range of industry adopters [2] even though it's
> still in v1alpha2.
>
> I would strongly encourage the community to contribute since the
> ease-of-use on distributed Kubernetes cluster is one perspective to compete
> with and learn from other frameworks. Kubeflow also has tf-operator and
> pytorch-operator which are being used by many companies.
>
> [0] https://github.com/kubeflow/mpi-operator
> [1] https://github.com/apache/incubator-mxnet/pull/17974
> [2]
> https://medium.com/kubeflow/introduction-to-kubeflow-mpi-operator-and-industry-adoption-296d5f2e6edc
>
> On Sun, Apr 5, 2020 at 3:10 PM sandeep krishnamurthy <
> sandeep.krishna98@gmail.com> wrote:
>
>> Thanks for bringing this here to MXNet community Yuan and thank you and
>> other contributors to bring MXNet operator in K8s.
>>
>> It would be very helpful if you can help us to by giving some details
>> about
>> current users of mxnet-operator in K8s or any users experimenting with it.
>> This info about userbase would help me and other committers to prioritize
>> this feature versus other feature requests in MXNet when we get proposals
>> by new members to contribute.
>>
>> Best,
>> Sandeep
>>
>> On Thu, 2 Apr 2020, 7:50 am Yuan Tang, <te...@gmail.com> wrote:
>>
>> > Hi community,
>> >
>> > We are currently looking into adding support for Kubeflow deployment of
>> > MXNet via the existing kubeflow/mxnet-operator:
>> > https://github.com/kubeflow/mxnet-operator. If you are interested in
>> > helping out and getting involved, please let us know in this issue:
>> > https://github.com/kubeflow/kubeflow/issues/4805.
>> >
>> > Cheers,
>> > Yuan
>> >
>> > --
>> > Yuan Tang
>> > https://terrytangyuan.github.io/about/
>> > <https://terrytangyuan.github.io/about/>
>> >
>>
>
>
> --
> Yuan Tang
> https://terrytangyuan.github.io/about/
> <https://terrytangyuan.github.io/about/>
>
--
Yuan Tang
https://terrytangyuan.github.io/about/
<https://terrytangyuan.github.io/about/>
Re: Kubeflow support for MXNet deployment
Posted by Yuan Tang <te...@gmail.com>.
Definitely. Thanks for asking and I am happy to provide more background.
The MXNet Operator has v1 controller. However, there has no formal release
on GitHub and it has not been included in Kubeflow by default yet. Though
users are welcome to use it as a standalone k8s operator. This will
hopefully be something the community will help drive forward as the initial
authors are not active. I don't have a good idea of who's using it yet
externally but I'll start by adding an ADOPTERS.md. *If your company uses
MXNet Operator, please let me know*. I will post the link to the PR once
it's ready.
Another easy-to-use option on k8s would be using Horovod + MXNet with *Kubeflow
MPI Operator* [0] (I just added a section in MXNet's doc on this [1]),
which already has wide range of industry adopters [2] even though it's
still in v1alpha2.
I would strongly encourage the community to contribute since the
ease-of-use on distributed Kubernetes cluster is one perspective to compete
with and learn from other frameworks. Kubeflow also has tf-operator and
pytorch-operator which are being used by many companies.
[0] https://github.com/kubeflow/mpi-operator
[1] https://github.com/apache/incubator-mxnet/pull/17974
[2]
https://medium.com/kubeflow/introduction-to-kubeflow-mpi-operator-and-industry-adoption-296d5f2e6edc
On Sun, Apr 5, 2020 at 3:10 PM sandeep krishnamurthy <
sandeep.krishna98@gmail.com> wrote:
> Thanks for bringing this here to MXNet community Yuan and thank you and
> other contributors to bring MXNet operator in K8s.
>
> It would be very helpful if you can help us to by giving some details about
> current users of mxnet-operator in K8s or any users experimenting with it.
> This info about userbase would help me and other committers to prioritize
> this feature versus other feature requests in MXNet when we get proposals
> by new members to contribute.
>
> Best,
> Sandeep
>
> On Thu, 2 Apr 2020, 7:50 am Yuan Tang, <te...@gmail.com> wrote:
>
> > Hi community,
> >
> > We are currently looking into adding support for Kubeflow deployment of
> > MXNet via the existing kubeflow/mxnet-operator:
> > https://github.com/kubeflow/mxnet-operator. If you are interested in
> > helping out and getting involved, please let us know in this issue:
> > https://github.com/kubeflow/kubeflow/issues/4805.
> >
> > Cheers,
> > Yuan
> >
> > --
> > Yuan Tang
> > https://terrytangyuan.github.io/about/
> > <https://terrytangyuan.github.io/about/>
> >
>
--
Yuan Tang
https://terrytangyuan.github.io/about/
<https://terrytangyuan.github.io/about/>
Re: Kubeflow support for MXNet deployment
Posted by sandeep krishnamurthy <sa...@gmail.com>.
Thanks for bringing this here to MXNet community Yuan and thank you and
other contributors to bring MXNet operator in K8s.
It would be very helpful if you can help us to by giving some details about
current users of mxnet-operator in K8s or any users experimenting with it.
This info about userbase would help me and other committers to prioritize
this feature versus other feature requests in MXNet when we get proposals
by new members to contribute.
Best,
Sandeep
On Thu, 2 Apr 2020, 7:50 am Yuan Tang, <te...@gmail.com> wrote:
> Hi community,
>
> We are currently looking into adding support for Kubeflow deployment of
> MXNet via the existing kubeflow/mxnet-operator:
> https://github.com/kubeflow/mxnet-operator. If you are interested in
> helping out and getting involved, please let us know in this issue:
> https://github.com/kubeflow/kubeflow/issues/4805.
>
> Cheers,
> Yuan
>
> --
> Yuan Tang
> https://terrytangyuan.github.io/about/
> <https://terrytangyuan.github.io/about/>
>