VMworld is wrapping up, so here's a look at some of what our team has seen in the last few days:
Not so turn-key VSAN appliances
One of the major value propositions behind converged and hyperconverged solutions is the notion that they are supposed to make virtualized infrastructure deployments vastly more simple. It would seem that at least some of the VSAN appliances on the market aren’t living up to the hyperconverged plug-and-play hype.
Case in point, the lead storage admin from Tribune Media test drove two different VSAN appliances and found that there was quite a bit of integration work that had to be done before they could be put to use. Turns out he opted to integrate VSAN himself using off-the-shelf commodity hardware. The good news is he was able to save money by taking a do-it-yourself approach, and he liked the fact that he had the flexibility to redeploy any of the components as their needs change. While it took some time for Tribune Media to stand up their VSAN environment, it has been performing well since going into production and is living up to their expectations thus far.
The takeaway – be sure to speak with several references before pulling the trigger on a hyperconverged solution. Many companies are claiming the mantle of hyperconverged simplicity and efficiency but not all are delivering on its promises.
VMware’s Project Photon is more than just VM-driven container management
As I’ve covered in two recent blogs, VMware’s Project Bonneville and Photon provide virtual administrators a way to satisfy the demands of their application developers to use containers, albeit within the envelope of a VM. The benefit is that Bonneville provides a “lightweight” VM that provides the isolation and security elements often needed in multi-tenant environments while allowing admins to use tools that they’re already familiar with, in this case vCenter. Photon is VMware’s longer term approach to managing containers and using container management frameworks like Docker and Kubernetes. VMware provided the analysts in attendance at VMWorld 2015 with a whiteboard glimpse into this architecture under the cover of a NDA. While I can’t divulge the details on what this future container management platform looks like, I can say that at least from an early design perspective, it is checking many of the must have boxes from a cloud native computing standpoint.
One of the fun things to do at VMworld is to view demonstrations of technology presented by the office of the CTO. These are cool innovative things being explored today, and although there’s no guarantee they will find their way into products, they do have a tendency to impact future product direction in some way. Concepts like Photon OS came out of the the CTO office. Ray O’Farrell, VMware CTO said that they have a pretty good track record of commercializing innovative ideas. Perhaps one can contrast that with PARC -- VMware’s neighbor across the street -- which has a history of great innovation, but didn’t have a great batting average in getting it successfully commercialized for Xerox, PARC’s original founder.
Here are a few demonstrations shown:
- Virtualized RDMA device. With this, remote direct memory access devices (RDMA) can be virtualized, just like virtualized NIC like a vmxnet device in standard vSphere. What this means is that ultra fast IO can be presented to guest operating systems, and it’s not a pass through of a device like a Mellanox interface card, but as a truly virtualized device. Making you yawn? What does it mean to you and me? Ultra-low latency apps, like voice processing, can run in a virtualized environment, and telco NFV apps can run at super fast speeds and full capabilities of features like vMotion can work on that VM. If you’re into this kind of stuff, it is amazing.
- Hybrid resource scheduling. This means that resource scheduling of VMs (like using DRS) can work across on-premises vSphere systems and off-premises (on the cloud). This means that it’s possible to place VMs on local clusters that can burst to the cloud, based on resource constraints on CPU, memory, etc. People are familiar with simple constraints today like CPU-affinity rules, but this can be much more powerful. This sort of technology can be used to really make a hybrid cloud transparent and useful.