On Mon, Sep 2, 2019 at 8:22 PM Yong Hu <yong.hu@intel.com> wrote:
Hi Felix, Please see my comments.
To let the community know you and your research or business, I would invite you join us into the weekly community meeting for a quick introduction about what you are doing/going to do with StarlingX.
https://wiki.openstack.org/wiki/Starlingx/Meetings#7am_Pacific_-_Community_C...
regards, Yong
On 02/09/2019 10:01 PM, Felix Kretschmer wrote:
Hi Yong, thanks. See my comments inline as well. Best, Felix
Am 2019-09-02 10:51, schrieb Yong Hu:
Hi Felex, pls see my comments inline. On 02/09/2019 4:27 PM, Felix Kretschmer wrote:
Hi Yong, thanks for the answer. Starting with OpenStack VMs. Containers via K8S might also be a later possibility. So everything running on OpenStack will work with STX? I know that windows works with OpenStack. YES, StarlingX doesn't change OpenStack. It's supposed to make deployment of OpenStack easier. The kvm in STX supports RT by VMs? Yes, it does. We ever tested a CentOS with RT kernel as the guest OS in VM.
Cool. Do you have any measurements (e.g. a cyclictest histogram) of this?
Yes, we have run cyclictest on the RT kernel of compute node and the result was very reasonable (in ms), but I am afraid I cannot disclose the data itself.
In the meantime, there is a performance test framework proposal (https://review.opendev.org/#/c/677287/) under review, in which cyclictest will be used as the RT test framework, and accordingly the test recipe will be presented as a part later. If you have more specific performance test requirements or even better you, as a community member, want to contribute into this area, you are definitely very welcome.
Hello Felix I will be more than happy to have a chat with you about the performance metrics you are interested in. Right now our list of tests cases that we are developing include: Detection of failed VM Detect failed compute node Controller Node fail detect/recover Network link fail detect DPDK Live Migrate Latency Vswitch: line rate 1-core Avg and Max host latency ( using cyclictest ) Avg and Max gst latency ( using cyclictest ) They will be part of the performance test framework that Yong mention before. In the end, you will be able to run the framework on your environment to get your personal metrics. Every environment is different and each performance result is affected by multiple variables, because of that, we encourage every user to use the incoming framework to get their own metrics. Regards
Is there any information about the demos you were talking about. Would be interested in gained experiences there. If you can articulate your use cases in details, we might share more about how to deploy.
My use case may be described as some kind of distributed setup of compute-nodes but they are filled with relative small workloads and therefore are not that rich in resources. So I might have a controller (with or dedicated storage is both possible at full hardware specs) and several nodes with hardware specs of: single socket CPU (quad or octa core), 8 to 16 GB (sometimes 32GB) memory and only 1 disk (HDD or SSD possible)
This kind of remote or distributed "small" compute node is also an ongoing task in community, you are welcome to join us, in terms of making requirement spec, use case spec and prototype development.
Best, Felix Von meinem Mobiltelefon gesendet
Am 02.09.2019 um 09:56 schrieb Yong Hu <yong.hu@intel.com>: Felix, pls see my comments inline. In addition, are you going to use OpenStack VMs or K8S containers? Regards, Yong
On 02/09/2019 3:24 PM, Felix Kretschmer wrote: Hello *, I am at the research of possible technologies for my use case and came across STX. Reading through the documentation and website I came up with a couple of questions and hope someone can help me by finding answers. 1. Realtime capabilities I did not find any information about the realtime capabilities. Is STX in the compute nodes ready for time deterministic workloads within VMs? e.g. PREEMPT_RT Linux or bare metal virtual machines? I guess kvm is the underlying hypervisor? Compute node host Linux could be with standard kernel or RT kernel. Similarly guest OS in VM can have one of them. Yes, we are using KVM and QEMU. 2. Windows VMs So far I did not find anything about Windows VMs. Due to the integration of Nova and the underlying technologies, I suppose this is possible without any doubts. Is it? Good question. We haven't tried Windows as guest OS in VMs. Technically it should be okay to install a Windows VM on top of OpenStack.Checking out this doc: https://docs.openstack.org/image-guide/windows-image.html 3. Hardware requirements Probably the most interessting topic for me, because the requirements for bare metal installations seem quite high (Dual-socket with 8 cores each and 32GB RAM) for compute nodes. Is this necessary? I wanted to deploy a couple of compute nodes, but with a lot smaller hw requirements as well (single socket octa or even quad cores with 16 or 8 GB of RAM). How much does the system of compute nodes itself needs? Quite a lot I suppose if the controller needs 64GB of RAM. It should be doable. In some experimental cases (like demos), we did install compute nodes on PCs (core based, 16GB RAM), though StarlingX preferable memory size is 32GB... As well, Supporting *smaller* work-notes ongoing effort... Thanks for any help. Best, Felix Von meinem Mobiltelefon gesendet _______________________________________________ Starlingx-discuss mailing list Starlingx-discuss@lists.starlingx.io http://lists.starlingx.io/cgi-bin/mailman/listinfo/starlingx-discuss
Starlingx-discuss mailing list Starlingx-discuss@lists.starlingx.io http://lists.starlingx.io/cgi-bin/mailman/listinfo/starlingx-discuss
_______________________________________________ Starlingx-discuss mailing list Starlingx-discuss@lists.starlingx.io http://lists.starlingx.io/cgi-bin/mailman/listinfo/starlingx-discuss