Ask Your Question

gomgom's profile - activity

2017-01-09 15:51:56 -0600 received badge  Notable Question (source)
2017-01-09 15:51:56 -0600 received badge  Popular Question (source)
2017-01-09 15:50:45 -0600 received badge  Popular Question (source)
2016-12-21 12:09:25 -0600 asked a question Automatic flavor selection based on available hardware?

Still trying to come up to speed on OpenStack, so the answer to this might be obvious...

Let's assume that an image has two flavors:

  • one based on specific hardware (specific PCI board), optimized, using low CPU core count and memory
  • one based on generic CPU resources, using a higher CPU core count, mem etc...

When instantiating this image, we always want to pick the hardware optimized one, to optimize resource use. But we want to be able to revert to the second image when the first choice is not available.

Is there a pre-defined way to go about this? It seems to me that the nova scheduler filter mechanism is built to answer to only one specific flavor request, but cannot arbitrate between two alternatives? Is it something an admin would have to handle outside Openstack?

2016-12-21 04:21:23 -0600 received badge  Enthusiast
2016-12-20 11:18:31 -0600 asked a question pci_passthrough, pci_alias and PCI subsystem vendor ID?

Hi,

I am working on a set of custom co-processing PCI boards (not networking cards). I need to identify systems hosting these boards, for resource allocation etc... The solution apparently relies on pci_passthrough, used as a property for the flavors I need to create.

One thing is that pci_passthrough seems to be able to rely only on the PCI Vendor and Device ID, but ignores the Subsystem fields. The problem being that it is not always desirable, or even possible, to change the Vendor and Device ID in the PCIe endpoint device. In fact, when designing a board relying on a 3rd party PCI IC, these fields should be left to the IC manufacturer's values. Only the Subsystem fields should be set to the board developer's custom values, reflecting the specific implementation for that IC.

Is there a way to use the Subsystem Vendor and Device ID fields in the pci_alias definition? I don't even know if libvirt exposes these fields...

2016-12-20 11:01:16 -0600 received badge  Scholar (source)
2016-12-13 09:16:51 -0600 received badge  Popular Question (source)
2016-12-12 12:13:26 -0600 commented answer PCI devices and enhanced platform awareness

Thanks a lot for the pointer, I'll be sure to read on that.

2016-12-12 09:26:52 -0600 asked a question PCI devices and enhanced platform awareness

Hello,

I am trying to come up to speed on OpenStack, working on a project involving custom PCIe devices. Part of what I'm trying to do seem to correspond to the enhanced platform awareness feature, described here: https://wiki.openstack.org/wiki/Enhanced-platform-awareness-pcie (https://wiki.openstack.org/wiki/Enhan...)

I am looking for ways to automatically detect PCIe device, and orchestrate instantiations based on PCIe devices present in a plant (my application running faster and requiring less resources when running with the custom PCIe board, and supporting hot swappable hardware).

When looking at this enhanced platform awareness blueprint, how do I tell if it got implemented (it seems to date from 2014)? Was it renamed to something else, or superseded by another OpenStack feature? Any documentation, discussion, developing members list anywhere?

2016-12-02 07:26:41 -0600 answered a question How to set up hardware preference for a given PCI board?

OK, I guess I found some elements today, but still don't have answers to my problem.

My understanding is that the resource assignment will be done by Nova. The filter ComputeCapabilitiesFilter can be extended to add pretty much any type of custom capability, as an extra_specs field. This filter could be used to select servers hosting custom hardware.

However I am still not clear on the following:

  • my understanding is that the extra_specs attribute of each host is set manually by the admin. I'm looking for an automatic way of detecting, and updating hardware status (to account for hot swappable boards), so that each server would publish their configuration on a regular basis
  • the filter mechanism can filter out hosts, and ensure that a VM is instantiated on the proper hardware. But what if my VM's requirements vary, depending on the presence of my PCI board? It is takes 1 core, 2G or RAM with the PCI board, and 4 cores, 16G without it? Is there a way of modelling this type of dual requirement?
2016-12-01 21:41:25 -0600 asked a question How to set up hardware preference for a given PCI board?

Hi,

I'm a complete newbie, so I hope this question is not completely irrelevant. I'm still learning about the high level concepts in OpenStack, but have not found what I'm looking for so far.

We have an application which is optimized when running on a platform hosting a custom PCIe board. However it can also run without it, at a much lesser speed.

Is there a way to specify a preference for a given hardware, specifically servers hosting a given PCIe board, when assigning a VM to a server? The PCIe board in question is not a standard one, it is not a NIC or a graphic card.

If so, is the PCIe configuration for each server defined statically when they're installed by the site admin, or is there a module which takes care of discovering a server's configuration, and publishing it to Nova or whichever element is going to schedule VM deployment in this case?

Thanks!