Cisco has finally decided to bring the vodka to spike the punch at the Hyper-Converged Infrastructure party. And it tastes pretty damn good. There have been rumours for a while now that Cisco was working with Springpath and as a major third round investor it’s not surprising to hear about their entrance into the HCI arena. The Register’s Chris Mellor reported about Something bubbling up at Springpath back in early December. So what is the offspring of Cisco and Springpath called? Cisco HyperFlex!!
Hyper-converged systems so far have delivered on simplicity and scale but there’s been a massive gap in the lack of network integration in existing solutions. Yes you can use top of rack fast switches. In some cases customers use Cumulus on whitebox top-of-rack switches for software defined networking but networking is not a built in feature of the two leading hyper-converged solutions, Nutanix and Simplivity.
HyperFlex joins the comprehensive DC portfolio along with UCS, MDS and Nexus. It means that Cisco now has a play in traditional component based infrastructure, converge infrastructure and now hyper-converged infrastructure. Cisco is adding HyperFlex to provide it with another string to its software defined infrastructure. It will now have:
- UCS – compute (service profiles, APIs etc.)
- ACI – for software defined network
- HyperFlex- software defined storage, compute and network
On the initial release Cisco HyperFlex will support file storage and VMware. There are a number of other storage types, such as block and object, and hypervisors on the roadmap. There’s also going to be container support. Given that Springpath was hypervisor agnostic I’d expect a quick ramp up from Cisco and fast feature release cycle.
Like pretty much every other hyper-converged solution Cisco sees its expected use-cases to be:
- Server virtualisation
- Test and development
- Large remote branch offices
UCS Manager is already familiar to multiple thousands of customers worldwide and the server and network deployment settings in HyperFlex come from pre-configured Service Profiles. Service Profiles are well and truly familiar to anyone that has worked with Cisco UCS. Given that customer base and the familiarity with existing management tools there’s massive potential for Cisco HyperFlex here. There are some well developed existing incumbents in the hyper-converged market with Nutanix leading the way and HyperFlex will allow Cisco to gain a foothold in that rapidly growing market.
HyperFlex requires 3 nodes to start a cluster, not unlike other hyper-converged systems. Springpath wrote their own file-system for the data platform which is what Cisco are using for HX Data Platform. The HX Data Platform sits as a management plane between the underlying hardware and the hypervisor. HyperFlex uses a log structured file system to create a stream of writes sequentially which improves both the performance and longevity of both flash and traditional hard drives. At present the write rates are unknown and it’s definitely something I’m keen to know more about. Dedupe and compression are inline tasks turned on by default and thanks to the log structured file system the performance impact is very low. HX Data Platform stripes data across all nodes simultaneously, leveraging cache across all SSDs for fast writes. It also provides balanced space utilisation with multiple copies of data throughout the cluster. This means that no data migration is required following a VM migration to another host.
HyperFlex can sustain 2 node failures and the replication rate can be dialed up to 3 copies of data to allow 3 node failures. If a node fails the evacuated VMs re-attach with no data movement required. The beauty of HyperFlex is that the replacement node is automatically configured via UCS Service Profiles. It also automatically re-distributes data to the new node. For engineers/admins with existing Cisco UCS knowledge this means an easy transition from existing UCS platforms to HyperFlex.
HyperFlex provides instant provisioning, cloning, and snapshotting of virtual machines from within vCenter (can spin up clones extremely quickly thanks to point based snapshots) and the clones and snapshots are pointer-based and space efficient. The creation and deletion of snapshots do not impact performance.
HyperFlex extends the existing Cisco UCS infrastructure and uses Fabric Interconnects to connect the nodes in the cluster. On release there will be 3 versions of the nodes available:
- HX200c – 1U – 3 to 8 node cluster (VDI & ROBO)
- HX240c – 2U – 3 to 8 node cluster (Server virtualisation, IT/Biz Apps, Test/Dev)
- HX240c and B200 Hybrid (CPU bound apps, VDI)
The HX Data Platform will be based on an annual subscription software model. Rough pricing has been released for the smaller HX smaller with 4 nodes and 2 Fabric Interconnect to come in at approx US$65,000. This puts it right in the line of other hyper-converged solutions and makes it a very interesting prospect. In order to start the cluster you need to buy 3 nodes and after that you can add individual nodes to the existing cluster. It is also possible to mix node types after the initial cluster setup, although this won’t be possible on first day of sales.
It is also possible to install IOVisor onto existing ESXi hosts and leverage storage from the HyperFlex so that you can scale the compute independently of your Hyperflex system and leverage your existing infrastructure and financial investments. This is something that’s not currently available to other hyper-converged players and it’ll be interesting to see how this part of the solution develops.
HyperFlex going to be primarily managed via vCenter thanks to plug-ins for the HX Data Platform. It will leverage service profiles in the Fabric Interconnects which will also link into vCenter to give one portal for everything. No new data interface or management consoles are required and this is key to Cisco’s entry into the hyper-converged platform as the knowledge already exists within Cisco’s customer base.
I think this move by Cisco is bold but it’s been one that’s been expected for some time now. They are obviously going to be gunning for the hyper-converged infrastructure leaders, Simplivity and Nutanix. I believe Nutanix is going to continue leading this field for a while yet but Simplivity will need to re-evaluate after this announcement. They already leverage Cisco rack servers in the form of their OmniStack solution. I think Cisco will still keep that relationship going, in a similar manner to the VCE relationship, but the collaboration between the two may suffer. Only time will tell.
This is an exciting time for Cisco and this is a great move for them. The HyperFlex solution extends their Data Center portfolio nicely and gives them a great platform to grow and add further features in the coming months. It’s definitely interesting times in the hyper-converged arena.