A bit about converged (and hyperconvergent) IT infrastructure

Data Center Ashburn (VA) / photo by Victorgrigas CC

Last week we tried to diversify Habr's tape with a couple of materials on our core topics:
')

• IaaS-digest: 30 materials on projects implemented in the cloud
• Experience of using IaaS by large (and not so) companies
• IaaS-cases: How the "cloud" helps the business

Today we decided to talk about convergent solutions that change the way IT services are delivered.

According to IDC, the global cost of converged infrastructure will reach $ 17.8 billion in 2016 and will amount to 12.8% of all costs for storage devices, network and software solutions. In 2012, this figure was $ 4.6 billion.

Convergent infrastructure

What is a converged infrastructure? This type of infrastructure is a ready-made solution from the manufacturer whose task is to accelerate the development of infrastructure. The basic idea is to provide one entry point for technical support and simplify component maintenance.

The term converged infrastructure was proposed by Hewlett-Packard six years ago. In Gartner terminology, this type of infrastructure is called an integrated system, and in Cisco Systems it is called Unified Computing System (UCS) (see Cisco UCS unboxing in our Habré blog ).

But whatever the name, behind it lies the same idea: combining memory, computing and network resources into a pool, pre-configured to work in a data center. This approach reduces the infrastructure sweep time from several months to several days.

Creating a converged infrastructure in a corporate environment is more than just replacing a pair of network devices. In a non-convergent infrastructure, implementing a virtual server, a hypervisor is launched on the physical servers that manages the virtual machines, and the data storage is a DAS, NAS or SAN. In a converged infrastructure, storage is combined with physical servers, and flash storage is used for high-performance applications and data caching.

An example of a converged infrastructure is a solution from Cisco and NetApp - FlexPod. Cisco is responsible for the server and network side (Nexus network equipment and UCS server equipment), and NetApp (FAS series) is responsible for data storage. The result was a flexible, scalable and fault-tolerant system.

The solution FlexPod uses the Swiss team Sauber Motorsport AG, which acts in the race "Formula 1". Back in 2007, the guys were faced with the need to organize a mobile data center in order to launch applications for analyzing race information, and then transfer the collected data to the storages of stationary data centers.

The company has deployed a two-node cluster consisting of NetApp FAS2040 and NetApp SyncMirror for data replication. The mobile data center contains 8 Cisco UCS blades and a Cisco Nexus switch.



The data obtained during the race using NetApp SnapMirror is replicated to MetroCluster in the data centers of Hinwil. The average delay in transferring data from the racetrack to the head office is about 15 minutes. Here is the cloud infrastructure of data centers in Hinwil:



With the help of the FlexPod platform, data on cars is collected. Information about the use of fuel, temperature, engines comes from hundreds of sensors. The team uses telemetry indicators along with road condition data, running simulations in FlexPod to compare how the virtual model relates to the current vehicle settings. This helps team engineers respond to changes during the course of the race.

Hyper-Converged Infrastructure

Hyper-converged systems take the concept of convergence to a new level. An important difference between the two technologies is that in a converged infrastructure, each component in the building block is discrete and can be used separately. As for the hyper-convergent infrastructure, it is a software-defined technology, because all the components are integrated.

Hyper-converged solutions are distinguished by improvements at the level of the program controller, which makes it easy to scale them. To increase capacity and performance, you need to add a new block. Instead of increasing power by increasing the number of disks, the amount of memory or processors, performance increases by adding more modules.

Thus, the hyper-convergent infrastructure is an infrastructure in which the computing power, storage, servers, and networks are combined into one whole using software tools, and they are managed through a common administration console. For this reason, instead of an IT professional team, it is often enough to have a single system administrator for managing data warehouses and server hardware.

Among the disadvantages of hyperconvergent systems, it is worth noting the impossibility of a granular upgrade. Increasing storage and improving performance are critical items for any company, but if space on the storage cluster comes to an end and computing resources are more than enough, you will still have to increase the total computing power by adding a new module.

A major supplier of hyper-converged infrastructure is VMware. VMware EVO: RAIL uses the Japanese bank Fukuoka Hibiki Shinkin. Atsushi Yoshida (Atsushi Yoshida), a representative of the bank, initially drew attention to EVO: RAIL due to its flexible integration with existing servers, data storage systems, the hypervisor and virtual software.



The use of a hyper-convergent solution allowed the bank to reduce operating costs by up to 10% and reduce the TCO (total cost of ownership) by 40%. The entire implementation time, taking into account the creation of virtual machines, took 15 minutes.

As with traditional infrastructures, the cost of the hyperconvergent infrastructure may vary depending on the hypervisor used. Infrastructure on VMware vSphere or Microsoft Hyper-V can be quite expensive, but Nutanix supports open-source KVM.

For those who are looking for a way to use already existing equipment to build a hyper-converged infrastructure, several companies offer software technologies that increase the performance of the existing infrastructure. One such company is Atlantis of Mountain View, whose solution turns the DAS into an array, increasing the number of VMs working with the storage.

Summarizing, we can say that convergent and hyperconvergent solutions offered by different companies help save time on implementation and technical support of the infrastructure; there is no need for a large number of administrators providing this support. A company that implements such solutions improves the efficiency of the IT department and its business.