[kata-dev] Summary of the Kata Meetup Beijing last week

Xu Wang xu at hyper.sh
Tue Oct 23 07:02:49 UTC 2018


Hi folks,
On Oct 17th Wed, we held a Kata deep dive meetup in Beijing. Most of the major
cloud providers, and ICT equipment vendors in China attended the meetup and
exchange their thoughts, practice, and use cases. [The participants](
https://etherpad.openstack.org/p/kata-meetup-beijing-2018) included Baidu,
Alibaba, Kingsoft, Tencent, QingCloud, Suning, IBM, UnionPay, Netease, China
Unicom, ZTE, Unicloud, AntFin, Lenovo, Spirent, etc. The one-day meetup was
supported by Intel (Sponsor), Huawei (Sponsor), Hyper.sh, and OpenStack
Foundation. I'd like to thank Maggie Liang from Intel especially, who
effectively led the organizing work of the meetup.
The executive director of OpenStack Foundation, Jonathan Bryce, who was visiting
China in the week, gave a short opening speech for the meetup. And before the
sessions, Yuntong Jin (Intel) and I (Xu Wang from Hyper.sh) gave an introduction
of the meetup and status update of the development in Kata Containers community.
In the morning, developers from Baidu, Huawei, and ZTE introduced their practice
on Edge, Public Cloud, and NFV cases. And Intel Developers gave an introduction
of NEMU VMM in the early afternoon. After all the sessions, the attendees were
grouped by topics and had a two-hours open discussion. The following topics are
addressed:
- In-sandbox networking policies: - Case:    - comes from Baidu;    - not a
kubernetes scenario;    - configure a network interface for the sandbox and
connect to the management network;    - group the processes in the sandbox to
two groups based on PID    - the processes from the provider could access local
(mgmt) network, while the other processes that running guest binaries don't have
the permission.  - Comments from Xu: we could generalize the requirement to
apply different networking rules for different processes in the sandbox;- VLAN
networking mechanism support and hotplug  - Case: - An NFV scenario;    -
Configure different VLAN for different tenants instead of overlays  - Comments
from Xu:    - The bridge + veth pair method might work for VLAN;    - The user
could write their own CNI plugin for accelerating;    - There is ongoing work in
the upstream community to help general CNI plugins to support Kata, however, it
might need to modify the current CNI interface and existing plug-ins;-
Networking performance metering and tuning:  - Many participants care about
networking performance;  - It's appreciated that all the users could provide
their real-life cases, which could help the community to find and fix the issues
better.
And there are many other topics are discussed as well:- 9PFS performance and
alternatives;- Block device based rootfs support;- Memory footprint
optimization;- Streaming media offload support;- Vsock support;- Container image
management issues for big container images (such as some tensorflow images are
8GB+).
Some of the feature requests or issue will be raised in Github.

Cheers,
--
Xu WangCTO & Cofounder, Hypergithub/twitter/wechat: @gnawuxhttp://hyper.sh
Hyper_: Make VM run like container
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.katacontainers.io/pipermail/kata-dev/attachments/20181023/e398d3e5/attachment.html>


More information about the kata-dev mailing list