Yesterday Cisco announced with much fanfare their entrance into the select club of hyper-convergence purveyors. In a market dominated by Nutanix and SimpliVity, with VMware VSAN and now also VCE VxRail every serious data center contender wants to have their place under the sun.
What is Cisco HyperFlex and what makes it specific? HyperFlex claims to be the ultimate solution to problems customers report with hyper-convergence and places itself as the all-in-one solution to these issues that it intends to resolve holistically – touching not only at the storage and compute, but also at the network fabric (and network intelligence).
From a hardware standpoint, no surprises, Cisco will be leveraging their UCS C220 and C240 rack models, under the HX 220c and HX 240c designations. The appliances look pretty smooth with a german Zeiss-like design. I did a check on the HX 220c M4 and the UCS C220 M4 and to details (type and size of SSD drives in bays 1 and 2, VIC 1227 card on the HX 220c M4) the specs are very similar.
In addition to this, and surprisingly enough for an HCI solution, Cisco allows the coupling of their HX 240c appliances with a Cisco UCS 5108 Chassis and UCS B200 M4 blades. This will allow for covering the imbalance of having more compute requirements than storage requirements, an interesting approach since generally speaking HCI solutions always boast about their linear scalability. In this case, we are able to have a compute to storage excess, but likely not the opposite. Maybe Cisco will throw in some Cisco UCS 6320 storage-heavy nodes sometime in the future?
I was not able to find numbers to how much this scales but considering the fact that FI 6248UP or 6296UP are part of the solution, we can guess that connectivity will likely not be the limiting factor here. Speaking about scalability, I’ve seen figures speaking of scaling up to 64 nodes in other blogs or articles, but nothing is confirmed.
The brains of the solution is of course hyper-converged storage (or do we dare say software-defined storage?). Software is eating the world, and sometimes our brains too. In any case the powerhouse behind Cisco HyperFlex is HALO from SpringPath. Did you know SpringPath before this announcement? Chances are high that you didn’t, and neither did I until 2-3 weeks ago.
What HALO promises: linear and non-linear scalability. For instance, a possibility is to leverage rack nodes for compute and storage capacity, and add blade nodes on top of that where only the caching layer is scaled across the entire cluster, meaning that persistent data will reside only on the rack nodes. In a similar way as what some other HCI vendors do, they leverage virtual appliances (one per node) that takes ownership of the storage components and create a distributed storage cluster. The data is striped across all of the nodes and a mechanism is in place to distribute or redistribute data in case of cluster imbalance (caused by adding or removing nodes, for instance). The number of replicas for a VM is also a customizable setting.
Standard features include deduplication and compression plus a native snapshot format that leverage VAAI, working with pointers which allows for very fast creation and deletion of snapshots, a very good feature IMHO.
A minimum of 3 nodes is required to create a cluster. Currently only VMware ESXi 5.5 / 6.x is supported with eventual additional support later on for KVM. In terms of management, they have a plugin that allows to manage the cluster directly from vCenter Server.
If you are interested in the mechanisms and innards of HALO, SpringPath presented a deep-dive of their platform at Storage Field Day 7 (see video below).
If you look for a shorter yet comprehensive enough review head to Cormac Hogan’s blog post here.
There was a discussion not long about the fact that with hyper-convergence you are displacing your storage backplane from the SAN fabric to the network infrastructure. When it comes to networking everybody knows how Cisco takes things seriously. What you get with an HyperFlex solution?
Seriously, you can order your HyperFlex solution with either a pair of Cisco UCS Fabric Interconnect 6248UP or FI 6296UP. Additionally, you could also purchase a pair of Nexus 9392 switches.
The unique competitive advantage of leveraging Fabric Interconnects is valid only if you intend to connect both rackmount and blade UCS nodes. I see this as an advantage for shops who already leverage UCS, FIs and eventually also Cisco ACI. For others, I think that it adds instead an unwanted layer of complexity. Regarding ACI, it seems to be optional and it has its pluses and minuses, which I already discussed in an earlier blog post from the TFDx series at Cisco Live. This makes however Cisco well-suited to deploy their HyperFlex solution in larger data centers.
HALO comes with good features such as the pointer snapshots and the decoupling of the caching and persistent tiers, allowing for a more flexible model than most HCI vendors provide. What I may have missed (maybe by lack of deep-diving?) is anything related to backups and DR.
I feel that the platform is coming at a time where the hyper-converged market is getting saturated. Contrarily to VMware (another late comer with VSAN 6.2), SpringPath does not benefits from VMware’s aura of trust and cannot leverage a huge installed base for upselling. Apart from the non-linear scale out (and the pointer snapshots) I do not see a feature that shines above others. Maybe their goal to become an all-encompassing hyper-converged platform supporting not only VMware but also other workloads (including physical) is a way to explore.
I am not fully convinced of Cisco’s statement that « other HCI solutions treat the network as an afterthought ». They certainly do not go to the depth and lengths Cisco goes, but let’s remind ourselves that nothing is for free either. A pragmatic analysis of the customer needs must be done prior to delve blindly into HyperFlex due to the port tax and potential costs is SFP+ is preferred over Twinax. I consider that HyperFlex is more an use case for large data centers than for smaller DCs and/or ROBO/manufacturing sites. The network investment seems to be too high for deployments under 5-10 nodes. With other HCI solutions you can simply leverage a pair of regular 10 GbE switches and Twinax and achieve cost savings, it really depends on the size of your deployment.
The promise of Software Defined Network is dependent on whether you will leverage Cisco ACI or not. I’ve written shortly about ACI in the past. I believe it has immense potential but is complex to integrate in brownfield sites. It comes down to how many HyperFlex nodes you intend to start with and how you expect to grow. From what I was able to ingest, it also seems that ACI is optional. On the other hand, if you are fully transitioning from a converged to an hyper-converged platform and want to build things from the ground up, it may be worth considering ACI from the very start of the project.
Cisco has gone to great lengths to present their solution and there was a torrent of quality blog posts + specialized press articles going out in a short timeframe after the announcement. Meanwhile SpringPath website remained desperately silent: no front page article, no news, no blog article (since then, they put a banner on their site). They do not seem very excited at the very least by the coverage their product is getting and the partnership they’re having. There has been some speculation about something bigger behind the « strategic partnership » announcement with Cisco since SpringPath went silent around November and cancelled several public appearances. This casts potential doubts on their future strategy and product updates. Update: during the time it took me to write this article, Chris Evans has come up with a very interesting article on the topic.
SimpliVity claims that it’s business as usual and they may very well be right. They have a proven, valid hyper-converged architecture, a well established customer base and a fruitful partnership with Cisco. In the first stages there’s not really much to fear from HyperFlex, but I may indeed be proved wrong.
Partners and customers alike will welcome HyperFlex with interest but also with caution. Every person who was treated with the Whiptail/Invicta marketing massage a couple years back will remember why. It is an audacious step for Cisco to partner with an hyper-converged platform that is not well known, not only because of the inherent risks of customer defiance towards this solution, but also in terms of timing.
What I find interesting, though, is that there are many customers around the world that love and trust the Cisco UCS platform. HyperFlex gives Cisco an unique opportunity to upsell their solution into already existing estates. That may give Cisco customers more confidence to jump into the hyper-converged world.
Time will tell if HyperFlex is here to stay, or if it’s an ephemeral « me too » attempt.