According to Gartner, and as you know they’re always right, by 2017 Web-Scale technology will be an architectural approach found operating in 50 percent of global enterprises, up from less than 10 percent in 2013! Large cloud providers like Amazon, Google and facebook, to name a few, basically reinvented the way their IT services are delivered. This on it’s own isn’t news, but now this technology, a.k.a. converged architectures, is available for ‘mainstream’ enterprise organisations as well. Small and big I might add.
Web-Scale technology in a nutshell
Itis, or should be, software defined making it extremely flexible, fault tolerant, scalable and able to run on standard x86 hardware. The idea behind Web-Scale infrastructures is conversion, a way to combine, or integrate, multiple infrastructural components like (server) compute, storage, networking and virtualization into a single platform or appliance of some sort. Resources are than aggregated and pooled, positively impacting performance, flexibility and overall efficiency. Management is centralized and simplified since converged infrastructures are, or should be, managed as a single entity no matter how big they get. Initial set up and configuration is a breeze, you should be able to have a basic infrastructure up and running within 30 to 45 minutes, if not, it’s simply no good. As a result maintenance should be minimal as well. It doesn’t come cheap (although what does?) but in the end all this will eventually lead to lower operational costs and ROI, you’ll see.
Scalability and white labels
Storage, which in most cases will include Flash or SSD like solutions, and networking (partly virtualized) will be ‘local’ to your VM’s increasing overall performance immensely, which is probably one of the biggest ‘eye catchers’. In terms of scalability IT will be able to scale / add what they need, when they need it a.k.a. scaling on demand, no more over-provisioning of storage, which is a big plus as well. As we all know, scalability is key!
Now this may sound as simple logic to some but if you look at most data centers today the above can be hard to find. Web Scale computing focuses on scale-out rather than scale-up technologies and since, as mentioned, its resources are aggregated directly from the underlying hardware, workloads can be scaled up without needing to scale up individual server and or other related hardware, again, offering simplicity and ease of management. As a side note on this, did you know that Google, in their datacenters, uses commodity, or white labelled, hardware which they build up themselves to keep costs low? And if Google can get away with it, who’s next?!
The power to act!
Today we have multiple vendors offering several Web Scale based solutions, and although one is slightly different than the other, in general, the all share some of the same characteristics, like the ability to address failing components, or any other faults and or errors that may arise in the underlying physical (converged) infrastructure. In most cases the software (remember that a Web Scale infrastructure will be software defined) is able to ‘act’ on these disruptions by automatically moving around workloads to another (least loaded and healthy) nodes, while in the meantime the faulty hardware can be replaced without any disruptions on the users side, it will automatically adapt to the situation and fixes it when needed!
Nutanix
Of course there’s a bit more to it than ‘just’ moving around some data, there’s a whole intelligence behind it. Web Scale uses technologies like, data deduplication, data tiering, writes are being replicated, physical components are redundant, multiple times over in most cases, data gets compressed, and the list goes on. All this combined make that Web Scale technology is extremely robust. Take Nutanix for example, you’ve must have heard about them by now, they offer something called software resilience, made possible by theircontroller VM’s, and I quote: All processes running in the CVM (the controller VM) are designed to fail fast when an error is encountered.
Components are continuously monitored, killed and restarted in the event of an error in order to recover as quickly as possible, rather than linger in non-responsive or corrupt state. Each host relies on its local Controller VM to service all storage requests. NDFS continuously monitors the health of all CVMs in the cluster. For example, if a Controller VM were to fail, Nutanix auto-pathing automatically re-routes requests from the host to a healthy Controller VM on another node. Check out Nutanix in my ‘sponsor’ section, they’re one of the leading companies when it comes to bringing Web Scale like technology into your datacenters. Just recently they became the first Web-Scale converged infrastructure vendor validated for Citrix XenDesktop! Another big step!
Let’s break tradition
Traditional storage networks (and we’re primarily talking SAN’s here) are complex, they consist of multiple components like switches, disk arrays and fibre optic HBA’s, they’re not that flexible in terms of scalability and often require a dedicated team of specialists when it comes to configuration, expansion and or troubleshooting. Support can be hard to get and support contracts are expensive. They’re not designed to handle today’s virtualized workloads and the (ever returning) IOPS that come with it although nowadays there are some ways (products) to boost performance (good ones like PernixData for example) but unfortunately it’s still far from ideal, something we’ll just have to deal with.
The same can be said for ‘traditional’ infrastructures, networking and server computing, they all require some sort of expertise and are all configured and managed separately from each other. Not a bad thing per se, but when things get complex (read big) it doesn’t help. And who are we to contradict companies like Amazon, Google, facebook and Microsoft who have all embraced these unified converged architectures as well? In fact, their architects basically invented the concept.
The best practices we once knew will soon become obsolete and known as ‘just’ practices, the way we once ‘did’ stuff. Another thing to think about is how we not only manage, but also need to support, upgrade and expand (scalability) our current infrastructures which isn’t an easy task with the hardware-centric architectures we’ve got going today.
It will take some time
I know, I get carried away sometimes, but you have to admit that there’s a truth in there somewhere, right? I’m not saying that SAN like solutions will disappear overnight, no way, it just wouldn’t be possible even if we’d want them to. And besides, despite some of their ‘handicaps’ they’re still very usable in most cases (I do like them, really) and way too expensive to just get rid off. But change is coming and it’s approaching fast!
For me personally, and as a Citrix engineer in particular, I think this is a good thing, we need more speed (massive IOPS and less latency), flexibility, scalability and ease of management. Fortunately there are companies like Pernix, Atlantis and of course Nutanix (to name a few) to help us overcome most of these challenges. Of course there will always be hardware involved, we need cooling, backup and power supplies, storage (disks and flash), electricity, networking components etc. It’s just the way that things are put together (converged) and managed (software) that is changing, drastically.
Bas van Kaam ©
Reference materials used: Nutanix.com, Gartner.com and wikipedia.com
One response to “What is Web-Scale technology and where does it come from?”
[…] The introduction of a new industry called Web-Scale technology with leading vendors like Nutanix. Bas van Kaam wrote an excellent article on WebScale technology. […]