Trend Hyper Convergence: Cisco HyperFlex
Modern enterprises are increasingly faced with tasks for which traditional servers and storage systems can no longer serve as an ideal solution. Combining the computing platform, virtual machines and storage system into a single distributed infrastructure allows the Cisco HyperFlex hyper-convergent infrastructure.
Since the advent of the first data centers, the main components of the data center have been the server farm, storage network and local area network. The latter combined the first two elements into a workable system.
After the spread of virtualization, data centers began to be a platform consisting of virtualized servers, network devices and storage systems. Since it became necessary to provide physical and virtual servers access to data equally, the requirements for storage systems have increased significantly, because now not a dozen servers could be connected to the storage system, but several hundred virtual machines at once. As a result, storage systems have evolved into virtualized hardware and software storage. In addition, it became necessary to virtualize switches to support different protocols, channels, subnets, priorities and speeds on one device. The resulting architecture combines various types of equipment, communication protocols and operating systems, called convergent.
')
However, hardware virtualized storage systems nevertheless continued to remain just a set of hardware - controllers and disk shelves connected to them, operating under the guidance of the operating system. Such an OS can be “packaged” into a virtual machine by running it on the main server, and the storage system itself can be placed on it as another service VM — multi-core and power of modern processors allows it. Moreover, such a storage system will provide not only the functionality of the storage system, but also allow organizing storage on a group of servers. Combining all this with a local network is also not difficult - 10 Gb / s Ethernet is now the standard for a corporate network, and many enterprises are beginning to implement 40 and 100 Gb / s links.
All this led to the emergence of so-called hyper-convergent solutions. Such platforms are no longer just a server, but a mini-data center in which resources are distributed between applications and users with a high degree of flexibility. Such a system is a Cisco Hyperflex. Let's see what she is.
What is in the package?
As already mentioned, the basis of the data center consists of three large subsystems: computing, switching and storage. As a fourth item, add management to this list. The first subsystem in HyperFlex is represented by UCS C220 or UCS C240 ​​servers. To increase computing power, it can be expanded through a chassis with blade servers; to increase storage space, add standard storage.
The network component includes a pair of switches based on the Nexus virtualized “converged” switch - UCS Fabric Interconnect 62xx. This solution is universal because it supports any protocol (IP / FC / FCoE / iSCS), in addition, it includes the Cisco UCS Manager management system with a free license, which provides extensive management capabilities for servers and their network part.
As for the storage system, the disks, the main element of any data storage, are located directly in the servers (nodes) that make up the platform. Storage is virtualized and has almost all the functions required by modern data warehousing in information infrastructures. At the same time, it is managed as one of the objects in the VMware vSphere virtualization environment. It is also worth noting that the VMware vSphere ESXi 6.0 hypervisor is already pre-installed on all HyperFlex nodes. In fact, to manage the entire system of an organization, it is enough to have an entry-level VMware specialist on staff.
As a result, by implementing the Cisco HyperFlex solution, the customer receives a complete data center out of the box, the configuration of which can be started from just three nodes. At the same time, the user immediately has a storage system distributed over the same nodes, which gives rather high performance indicators according to test results. Important is the fact that such a virtualized cluster requires the organization of a 10 or 40-gigabit local area network, which does not always exist in the infrastructure of organizations. A pair of universal UCS Fabric interconnect switches ensure high performance of the network part of the solution and correct operation of the distributed storage system.
Scaling features
However, with today's dynamic development of the IT industry, sooner or later, the company will face a lack of resources and the need to scale the existing infrastructure. In the case of Cisco HyperFlex, everything is quite simple.
With a shortage of computing resources, you just need to purchase additional memory up to 768 GB - this is 24 slots in both HX220s and HX240s. You can also connect the UCS5108 chassis, which makes it possible to get another 8 blades in the general pool of resources. The problem of lack of disk space is solved by installing another HX240c node to 24 disks, which will simultaneously increase the disk space, memory capacity, and the number of processors for virtual machines. That is, it is possible to increase the solution both in terms of critical characteristics and in a complex way, expanding the cluster. The latter option is particularly convenient, since new nodes arrive at the customer’s site already in a prepared form and the existing HyperFlex cluster automatically recognizes the new node and puts it into operation. No special action is required from the solution administrator.
In the case when virtual machines require a lot of space or they have the task to make backups to old hardware storage, you can connect classic storage systems of any manufacturer using 1- or 10-gigabit NFS / CIFS / iSCSI / protocol to a couple of converged UCS Fabric Interconnet switches. FCoE or Fiber Channel.
Basis for data center
In the event that just “a data center out of the box” is no longer enough, and a full-fledged data center with classic infrastructures is needed, Cisco HyperFlex also allows you to save investments - after all, a number of necessary components are already available. For example, the nodes of the HX2XX series are the same UCS-C-series servers, which are easily integrated into the data center. Nexus 6200 interconnects are also used in a full-sized data center, while the UCS Manager allows servicing up to 160 physical servers and / or storage systems with any connection protocol.
The disk space located on the virtual storage as part of HyperFlex may be available to other platforms.
In addition, you should not forget about such a factor as free space in the server, which is not always possible to expand. For installation in an existing rack, the blade chassis will also have to work hard by swapping equipment and cabling. And for the implementation of HyperFlex in the rack, you need only 5 units - one for each of the three nodes and two for interconnects. An even more serious decision with the HX 240c nodes will require only 8 RU. It is worth noting that the units do not have to be placed in a row - the main thing is to connect the nodes to the interconnect factories, and the nodes themselves can be “scattered” in the data center, taking 1 or 2 U where there is space.
As a conclusion
Based on the foregoing, Cisco HyperFlex will be a good choice for companies starting to implement an integrated IT infrastructure. Especially this solution is worth paying attention to organizations that are planning to implement as soon as possible and at the same time do not have a huge IT department. Installing HyperFlex on just 3 servers allows you to organize a full 10 Gigabit IP and FC infrastructure and provide data center management. The system will become the core of the data center, including through the capabilities of Fabric Interconnect.
Since the advent of the first data centers, the main components of the data center have been the server farm, storage network and local area network. The latter combined the first two elements into a workable system.
After the spread of virtualization, data centers began to be a platform consisting of virtualized servers, network devices and storage systems. Since it became necessary to provide physical and virtual servers access to data equally, the requirements for storage systems have increased significantly, because now not a dozen servers could be connected to the storage system, but several hundred virtual machines at once. As a result, storage systems have evolved into virtualized hardware and software storage. In addition, it became necessary to virtualize switches to support different protocols, channels, subnets, priorities and speeds on one device. The resulting architecture combines various types of equipment, communication protocols and operating systems, called convergent.
')
However, hardware virtualized storage systems nevertheless continued to remain just a set of hardware - controllers and disk shelves connected to them, operating under the guidance of the operating system. Such an OS can be “packaged” into a virtual machine by running it on the main server, and the storage system itself can be placed on it as another service VM — multi-core and power of modern processors allows it. Moreover, such a storage system will provide not only the functionality of the storage system, but also allow organizing storage on a group of servers. Combining all this with a local network is also not difficult - 10 Gb / s Ethernet is now the standard for a corporate network, and many enterprises are beginning to implement 40 and 100 Gb / s links.
All this led to the emergence of so-called hyper-convergent solutions. Such platforms are no longer just a server, but a mini-data center in which resources are distributed between applications and users with a high degree of flexibility. Such a system is a Cisco Hyperflex. Let's see what she is.
What is in the package?
As already mentioned, the basis of the data center consists of three large subsystems: computing, switching and storage. As a fourth item, add management to this list. The first subsystem in HyperFlex is represented by UCS C220 or UCS C240 ​​servers. To increase computing power, it can be expanded through a chassis with blade servers; to increase storage space, add standard storage.
The network component includes a pair of switches based on the Nexus virtualized “converged” switch - UCS Fabric Interconnect 62xx. This solution is universal because it supports any protocol (IP / FC / FCoE / iSCS), in addition, it includes the Cisco UCS Manager management system with a free license, which provides extensive management capabilities for servers and their network part.
As for the storage system, the disks, the main element of any data storage, are located directly in the servers (nodes) that make up the platform. Storage is virtualized and has almost all the functions required by modern data warehousing in information infrastructures. At the same time, it is managed as one of the objects in the VMware vSphere virtualization environment. It is also worth noting that the VMware vSphere ESXi 6.0 hypervisor is already pre-installed on all HyperFlex nodes. In fact, to manage the entire system of an organization, it is enough to have an entry-level VMware specialist on staff.
As a result, by implementing the Cisco HyperFlex solution, the customer receives a complete data center out of the box, the configuration of which can be started from just three nodes. At the same time, the user immediately has a storage system distributed over the same nodes, which gives rather high performance indicators according to test results. Important is the fact that such a virtualized cluster requires the organization of a 10 or 40-gigabit local area network, which does not always exist in the infrastructure of organizations. A pair of universal UCS Fabric interconnect switches ensure high performance of the network part of the solution and correct operation of the distributed storage system.
Scaling features
However, with today's dynamic development of the IT industry, sooner or later, the company will face a lack of resources and the need to scale the existing infrastructure. In the case of Cisco HyperFlex, everything is quite simple.
With a shortage of computing resources, you just need to purchase additional memory up to 768 GB - this is 24 slots in both HX220s and HX240s. You can also connect the UCS5108 chassis, which makes it possible to get another 8 blades in the general pool of resources. The problem of lack of disk space is solved by installing another HX240c node to 24 disks, which will simultaneously increase the disk space, memory capacity, and the number of processors for virtual machines. That is, it is possible to increase the solution both in terms of critical characteristics and in a complex way, expanding the cluster. The latter option is particularly convenient, since new nodes arrive at the customer’s site already in a prepared form and the existing HyperFlex cluster automatically recognizes the new node and puts it into operation. No special action is required from the solution administrator.
In the case when virtual machines require a lot of space or they have the task to make backups to old hardware storage, you can connect classic storage systems of any manufacturer using 1- or 10-gigabit NFS / CIFS / iSCSI / protocol to a couple of converged UCS Fabric Interconnet switches. FCoE or Fiber Channel.
Basis for data center
In the event that just “a data center out of the box” is no longer enough, and a full-fledged data center with classic infrastructures is needed, Cisco HyperFlex also allows you to save investments - after all, a number of necessary components are already available. For example, the nodes of the HX2XX series are the same UCS-C-series servers, which are easily integrated into the data center. Nexus 6200 interconnects are also used in a full-sized data center, while the UCS Manager allows servicing up to 160 physical servers and / or storage systems with any connection protocol.
The disk space located on the virtual storage as part of HyperFlex may be available to other platforms.
In addition, you should not forget about such a factor as free space in the server, which is not always possible to expand. For installation in an existing rack, the blade chassis will also have to work hard by swapping equipment and cabling. And for the implementation of HyperFlex in the rack, you need only 5 units - one for each of the three nodes and two for interconnects. An even more serious decision with the HX 240c nodes will require only 8 RU. It is worth noting that the units do not have to be placed in a row - the main thing is to connect the nodes to the interconnect factories, and the nodes themselves can be “scattered” in the data center, taking 1 or 2 U where there is space.
As a conclusion
Based on the foregoing, Cisco HyperFlex will be a good choice for companies starting to implement an integrated IT infrastructure. Especially this solution is worth paying attention to organizations that are planning to implement as soon as possible and at the same time do not have a huge IT department. Installing HyperFlex on just 3 servers allows you to organize a full 10 Gigabit IP and FC infrastructure and provide data center management. The system will become the core of the data center, including through the capabilities of Fabric Interconnect.
Source: https://habr.com/ru/post/327550/
All Articles