Virtualisation is fine, but may not deliver all the required benefits in managed services, where the whole point is automation, not another layer to configure. UK-based managed services provider Amatis has just invested in Tintri VMstore to underpin its virtual environment.
The application-aware solution has cut the MSPs data centre footprint and power consumption by 66% and reduced latency in hosted customer environments by 98%, says Alex Fossa, CTO and founder of Amatis. Amatis delivers a range of hosted services, including private and hybrid clouds, application hosting, collaboration solutions, and offsite data storage.
He spoke to IT Europa about the current issues in managed services and how customer requirements have changed in the last few years, “I think where we come from as a business supplying the corporate world is the focus on speed and even more on latency. So with virtualisation you split up servers, and load balance between them, so when you have to send large amounts of data between them, you need speed and no delay. Which is something we had always had before upgrading to Tintri.”
Amatis is a merger of a network services and managed services business; “We concentrate on infrastructure for app delivery- as a channel only business, we don't get involved in end-user applications.” Business is good, and expanding, he says. “We have international customers already with data in the UK and a private interconnect. Later this year we will offer France and Germany for our customers with European requirements, and guarantee the bandwidth all the way back to our data centres.”
“For the type of customer we deal with, the most important thing is not cost, but availability and performance. Everybody knows you get what you pay for, taking a cheap shared hosting route will mean issues, whereas we are providing an infrastructure platform for people to run applications; the customers need to know about guaranteed resource and response times, with some form of HA backup. Most people will have experienced a failure of some sort. We own and operate two data centres, so we can give them the comfort of knowing where their data is.
“We looked into it and it was the self-adapting service that meant we would not have to manage it. It saves a lot of admin time. I did a presentation not long ago where I spoke about the cost of IT and cost of virtualisation. There is misconception that users can turn their server rooms with 50 units into four and save a lot of money with virtualisation but all you are doing is reducing the power load and while there are other advantages, you still have to manage it and there is this layer sort of 'trowelled onto' the virtualisation which is another overhead. The only way to make the cost of virtualisation work is to cut this management overhead, so putting something like Tintri in to manage the 5%-10% of Opex which is storage management costs, works to bring this close to zero.”
Because Tintri can self-manage and is VM-aware, seeing the virtual machines, it can adapt policies on individual virtual machines, avoiding delays. “And we don't have the effect of one machine swamping all the resource, and causing a 'noisy neighbour' effects on others,” he says.
There had been a tendency to throw more storage in without worrying about the management, he agrees. Ten years ago, a database project, for example, would mean buying servers and a load of drives, and dedicating them to the project. Now it is virtualised, but still needs looking after.
Latency and speed issues matter, he says. Customers are now defining managed services requirements more than they used to, he says. A few years ago, customers were not really sure what they wanted, just looking for an easy way to offload in-house workloads. Now they are more application specific and the apps themselves are supported on the environments where they never used to be. Vendors such as IP telephony are supporting their products on public and private cloud environments.
Customers want environments where they still keep some of their data in-house for whatever reason. “They want to copy and replicate between in-house and outside. Most of the customers we have are in private cloud because they need the security and performance.”
The big thing that I feel is happening in the networking provider industry is that everything is movings towards software defined networking SDN, where, although all the main vendors have an offering, it has not seen traction. We are questioned on it, but no customer is really pushing for it. What will start happening is that ISPs will start doing more on a virtualised environments – firewalls, load balancing and routers, so there will be less need to buy tin with the software on it, or appliances in a data centre. It will be software as a VM image and installed that way. The intellectual property will be in software, and will be more efficient to install and manage, which is the point I made at the start.”
But it relies on a few key things: the environment running it has to be dedicated and managed, of top quality without cutting corners, the system needs to be low latency, with low latency storage infrastructure, and very good, reliable computing infrastructure - and it all needs to be scalable and easy to expand. So we use Cisco UCS servers where we can just slot more blades in as we need more resource, Cisco Nexus switching to ensure it all operates at low latency, and Tintri to simply manage the storage.”