Welcome back to part two of our blog series discussing how we came to our decision about which storage solution was best for Evolution Recruitment Solutions, and why. If you missed part one, you can read it here. If you are already up to speed, let’s get back to it!

Solutions

So, with our requirements and considerations defined as mentioned in part one, we looked at a number of different approaches. Could some sort of cloud or hybrid cloud offer a solution? What about hyper convergence, or converged? Should we stick with the traditional SAN type of appliance, or a DIY type solution?

The first of these to be ruled out was a cloud or hybrid cloud solution. We have connected our datacentre via IPSEC VPN tunnels to Amazon AWS and Microsoft Azure, and we do make good use of the flexibility and scalability of both of these cloud service providers. However, where the industry is in terms of its evolution of businesses migrating to the cloud, and where Evolution is on this journey right now, cloud solutions weren’t a viable option.

Everything, in terms of infrastructure and the business services, would need to be re-architected, redesigned, and migrated in order for it to be viable via one of these cloud providers in a cost-effective way. Furthermore, resource would need to be skilled up or hired. Migrating to the cloud is a completely different topic of conversation, and I could go into a lot more detail, but since this is about storage, let’s try to stay on topic.

The DIY approach?

Next, we ruled out the DIY approach. We could definitely build our own rack mountable servers, pack them full of storage and make them into a storage cluster. This would be fast, highly resilient and offer the flexibility to expand easily – the geek in me really likes the idea! Getting hands on, building our own bespoke storage solution, and we could definitely do this far cheaper than any of the other approaches. However, if we have any problems, who would be able to help us?

Hardware would be from numerous vendors, and you might get one vendor blaming the other for the cause of any problems. A vendor may pull support without any prior notice when used in conjunction with another vendor’s product, and you might not even discover this until you’re trying to troubleshoot a production down issue. Firmware upgrades would be entirely our own responsibility and would be risky, as it potentially may never have been tested before in that environment, and we don’t have a test environment.

Monitoring of the solution would, again, also be our own responsibility. Considering the storage solution was going to house the entire business infrastructure, and I was going to be accountable for delivering the maximum uptime of business services, this approach introduced far too much risk than I would be comfortable with. So, for these reasons, I ruled this approach out.

Hyper-convergence?

We only really reviewed two different hyper-converged solutions; Nutanix and SimpliVity. Both are amazing pieces of engineering, offering everything that we wanted and more. There is a lot of hype around this state-of-the-art technology, and it is certainly at the forefront in terms of the natural evolution of datacentre technology and its hardware uptake.

Businesses and organisations are moving away from purchasing separate servers, storage and networking, and everything that’s involved with building and managing those environments. These new hyper-converged products put all the servers, storage and networking, scaled down, into a few rack units of space. The products have been designed and engineered so all the parts work together in one chassis, you only have to contact one vendor for support, and they consume less power, so your energy bills and carbon footprint are reduced. Unfortunately for Evolution, there were two clear problems; the first being the cost.

Both of these products are expensive; around £50,000 just for one. The cost of Nutanix was heavily discounted for being a new customer, but the initial discount only applies to the first node purchased, so scaling out, which is one of the biggest selling points, would see around a 70% price increase. In both cases, a single node was a hybrid storage array, not all flash, and the servers gave us less RAM than the servers we are currently using at the moment. This lead me to our second problem; our existing servers and networking have years left on their warranties. The business has made investment in these areas not that long ago, so what are we going to do with the existing hardware?

These products were offering us more compute and RAM, which we don’t really need, and our servers, storage and networking don’t require a refresh all at the same time. This is a key point where hyper-convergence becomes applicable. If you have entire clusters of servers, networking and storage which be EOL altogether every 5 years, then hyper-converged is a perfect fit; but not in Evolution’s case.

Further solutions we considered

We also reviewed two converged solutions, EMC ScaleIO and vSAN, before they had been acquisitioned by Dell which now produces ReadyNodes. Both follow a similar concept to the hyper-converged solutions and are software-defined storage architecture solutions; they are very similar. The solutions involve making use of our existing servers and networking, by simply installing physical storage into the existing servers, then using software to make all the storage in each server available to the cluster. This provides highly available, resilient shared storage, which delivers fantastic performance.

Unfortunately, all our servers are Dell PowerEdge R730s with dual SD cards for internal storage. The servers have not been built to house any physical disks, and they do not have any RAID controllers. The chassis of the servers would need to be adapted, and additional hardware would need to be purchased to cater for these solutions, meaning the overall cost would go up significantly.

We priced up 12TB of All-Flash storage, which would provide us with about the same amount of capacity as the existing Dell EqualLogic, and with the hardware and software license costs, this amounted to around £27,000. Although you can implement these solutions with three servers, it is recommended best practise that you implement these solutions with at least four servers in each cluster. The VMware license for our server cluster is an Essentials Plus license, which limits us to a maximum of three hosts, and it has also been known for hardware to be removed from the HCL in the past, meaning you could build the solution, only to find out later that it is no longer supported. Obviously, if you buy the new ReadyNodes offered by Dell this isn’t an issue, but at the time we reviewed these solutions, Dell hadn’t bought VMware or produced the ReadyNodes. This is why these solutions were not an ideal fit for Evolution.

Looking at the traditional SAN approach, we reviewed the Dell EMC Unity 300, Dell Compellent SC4020, Tintri T820, Nimble CS235, Quantum QXS-4 series and Pure Storage M10 series. All were hybrid arrays apart from two of them, Pure Storage and Dell Compellent, which were both flash. Typically, the hybrid arrays were cheaper and provided more overall storage capacity. Most came with three years’ warranty, but would typically provide a maximum of seven years warranty, with the exceptions being Pure Storage and Nimble, which both offered no EOL for their products as they provided options around upgrading the controllers.

The options to expend future storage capacity varied; some required another node purchased, some required adding a shelf, and some came with empty drive bays which could be populated in the future as required – the latter being the most cost effective for future expansion. All have dial home proactive support, and in terms of functionality, reporting and monitoring directly from the appliance, and what the SAN offered overall, I felt Pure Storage, Tinitri and Nimble were the best, and probably in that order. Pure Storage was the most expensive by quite a way, due to their support and maintenance costs, and Tintri was second which was also a hybrid array.

The following table shows how they lined up against one another:

TBC…

Don’t forget to look out next week for blog number three, the last in this blog series, to find out about our final decision on which storage solution we decided on for Evolution!