Thanks to everyone who joined SDNCentral’s latest DemoFriday with Wind River as they showcased how high reliability and availability in a multi-vendor environment helps NFV deployments. The presentation included a demo of Wind River’s Titanium Server – a carrier-grade software solution – and also included a special proof of concept created by Wind River, Brocade, and HP.
After the demo, our presenters were nice enough to take some questions from our participants. Below, you can read the Q&A, watch the full presentation, or check out the teaser video and other exclusive resources.
Does Wind River provide individual, standalone components of Titanium Server, or just the turnkey binary?
Wind River: Wind River sells Titanium Server as a complete turnkey system. Titanium Server is made up of a number of components, which are also individual products available for sale.
Does Titanium Server only run on HP rack mount servers?
Wind River: Titanium Server has been certified on HP servers but will also run on others. For example, Kontron and Artesyn are both members of the Titanium Cloud Ecosystem which offers pre-validated integration.
Regarding “reliable” operation: during a server switchover, is there a (however brief) momentary disruption in service until the standby comes up? If so, what is the maximum time disruption that would still count as “continuous” operation vs “brief outage”?
Wind River: Assuming the question’s referring to a planned live migration. In such a case the maximum outage duration is 200ms (deterministically – independent of traffic).
What’s the routing strategy for Titanium server? (i.e. how do different tenant networks communicate and how do tenant networks communicate with external networks?)
Wind River: Titanium Server uses VLANs for tenant network separation and VXLAN for communication across networks. Tenant networks can also be administratively provisioned for shared communications if so desired.
Regarding “surge”, what is the advantage of spinning a 2nd VM on the same physical processor vs increasing the resources allowed to the existing VM?
Wind River: Under real world conditions, a new VNF would typically be launched on a separate physical server. While scaling up (vs. scaling out) is a reasonable alternative, it was not included in the initial PoC, but may be added in the future.
If we have a series of remotely monitored virtual machines, and one of them crashes, but doesn’t reboot, what are the recovery operations, and how is hardware insured?
Wind River: One of the functions provided by Titanium Server is to monitor VNF health, either via externally visible events, or using a client heart facility (as discussed during the demo). Should a VNF become unhealthy, it would be restarted by Titanium Server.
Regarding the question on hardware health, Titanium Server actively monitors underlying server health, alarming faults as found, migrating VNFs where necessary and initiating hardware recover if possible.
Have you simulated 2 problems at the same time? High traffic with a vRouter failure for example.
Wind River: Not specifically during this PoC, but multiple faults are tested as part of the product development process. The suggestion is a good add to a future revision of the PoC.
In the scaling demo, when does the newly created instance shutdown? Immediately after the CPU threshold returns to normal or after it remains stable for a certain period? Is this shut down duration configurable?
Wind River: The shutdown period is completely configurable at the time the service chain is defined; significant flexibility is provided to the operator. Further, Titanium Server provides a mechanism for VNFs to defer or refuse the shutdown request if circumstances make it inadvisable to the VNF. This is an optional capability.
Is the 200ms migration pause maintained under any traffic load or is it only for MTC windows where there’s minimal traffic?
Wind River: Yes, see response to question #7.
Can these use cases be done using OpenStack?
Wind River: Titanium Server builds on and leverages OpenStack as part of its implementation.While vanilla OpenStack provides base capabilities, it does not provide the reliability and automated healing capabilities of Titanium Server as demonstrated in this PoC.
Is there a graphical interface where we can see equipment status?
Wind River: Yes, the Inventory and Hypervisor panels of the Titanium Server GUI provide such information.
Where/how are the lost packages tracked during the node/vm recovering? Are they recoverable?
Wind River: When a compute node completely fails, the VNFs residing on it are lost and are automatically re-instantiated on another compute node by Titanium Server. Typically, VNFs implement their own resiliency mechanisms (active/standby, active/active, etc.) to deal with the case of an instance failure. Such a mechanism was demonstrated with the PoC’s use of the VRRP protocol. If the VNF is completely simplex (unusual in a telecom environment), then the active context it was working would be lost.
How is the load monitored on VMs?
Wind River: Titanium Server monitors a wide variety of VM load indicators through the kernel and Hypervisor: CPU occupancy.
Which VIM is supported by Wind River?
Wind River: Titanium Server is itself a VIM (Virtualized Infrastructure Manager).
What techniques do you use to get a high level of interrupt response determinism?
Wind River: Please contact Wind River for details on our Open Virtualization, KVM product and the kernel enhancements which it includes. This is a complex topic which requires a deep dive in its own right.