[Openstack-operators] ServiceVM/Device manager project useful for operators?
Yamahata, Isaku
isaku.yamahata at intel.com
Tue Jun 3 04:18:17 UTC 2014
Hello OpenStack Operators!
I'm wondering we want to tackle ServiceVM/Device life cycle management
[1][2] because I am NOT SURE if the framework is valuable to
operators/users. Here ServiceVM means VM that runs (network) services.
Your opinion will help us in determining whether to invest in
developing ServiceVM life cycle management as a separate project.
It would inform vendors that operators may find this of value and thus
to get involved in developing the same.
Would you see value to a framework that supports deploying services in VMs
such as firewall, load balancers, routers, switches etc? A uniform
way to deploy and manage such services? Lowers barrier to entry for
vendors to deploy their solutions to OpenStack/cloud
How are these different from other VMs?
1) Network related
(As we started from Neutorn networking project, so it's biased for
networking, but the framework for lifecycle management is
generic. not specific to networking.)
2) Categories such as firewall, intrusion detection, load balancing,
switch, router ..
3) Multiple vendor solutions (plugin/driver architecture)
4) Vendor specific management dashboards
5) Configuration
It will use neutron API for service chaining and traffic steering,
nova APIs to launch VMs. basically it re-uses as many openstack
services as possible.
This service would also correspond to VFN manager/catalog in NFV mano
architecture and related interfaces.
Development will be addressed as phased approach, something like:
1. Service catalog: could demonstrate usefulness in that the neutron
flavor api could make a call to our service catalog.
2. define/implement public Rest API/data model
3. VM/service scheduling and configure services with PoC use case
(haproxy, iptables in linux VM)
At this phase, the management inteface will be connected to
openstack management network with the assumption that VMs/services can
be trusted.
At this phase. actual VM/service could be spin up/down and deployed.
4. support side communication channel for VM/service configuration
enhance oslo.messaging with Marconi as discussed in Atlanta.
With this supported, unreliable VMs/services can be used.
5. actual horizon(GUI) support
architecture block diagram:
+------------------------------------------------------------------------+
| servicevm/device manager service |
++------------------++------------------------++------------------------++
|| || || ||
||Service catalog ||Service Managers ||Infrastructure ||
|| .Vendor || .lifecycle(Nova,Neutron|| . Mgmt interface ||
|| .Service type || .Configuration || . communication support||
|| .Image(In Glance)|| .Events/alerts/logs || . security ||
|| .Managemnet || || ||
|| .HW-flavor ++------------------------++------------------------++
|| | |
++------------------+ |
| |
+------------------------------------------------------------------------+
|
|
v
+-------+ +-------------+
|Horizon| |NFVC instance|
+-----+-+ +------+------+
+-+--+ +------+------+
|NOVA| |VNFC instance|
+--+-+ ------> +------+------+
+-+-----+ +------+------+
|Neutron| |VNFC instance|
+-------+ +-------------+
Horizon & CI: invoke service catalog APIs
Nova: VM spin up/down, VM scheduling
Neutron: service chaining/insert, service scheduling
References:
[1] https://wiki.openstack.org/wiki/ServiceVM/Incugbation
This page is draft under review
[2] https://wiki.openstack.org/wiki/Meetings/ServiceVM
[3] https://blueprints.launchpad.net/neutron/+spec/adv-services-in-vms
[4] https://docs.google.com/presentation/d/1gmuBprzedbvTQBt98JYiDFgLlTipDjr1ZgaEGL0xkcs
Regards,
Isaku Yamahata
More information about the OpenStack-operators
mailing list