IT budgets in organizations typically do not grow as fast as the requirements for the speed of deploying new applications and expanding existing ones. The solution is to redirect attention and human resources from the details of data center implementation to the applications themselves, which are closer to business. This can be done in several ways, and a public cloud that saves costs (not only) through economies of scale can be one of them.
When you look for the number and volume of technology company acquisitions, Cisco ranks second in both categories, number one in volume (1991-2018). Not surprisingly, it is often possible to hear the misconception that acquisitions are the only way for Cisco to secure evolutionary benefits in technology armaments.
The "California" or "Unified Computing System" (Cisco UCS) project is a product of Cisco's rich internal development. Hyperflex is another stage of development of the new architecture, which began in 2009 and completely changed the architecture of data centers cisco one data center.
The concept of architecture is crucial here, because UCS is not "another server from Cisco", but a system based on the idea of "Stateless computing". It abstracts the configuration of the server and its LAN and SAN connectivity from the hardware. The content of the term server in this case is not something physical, but a complete definition of the configuration, including unique identifiers such as MAC, WWPN, UUID, zoning and others in the tool "UCS Manager".
UCS Manager is home to flash memory in a device called Fabric Interconnect, which is not only the center of management for up to 20 chassis or 160 servers, but also the central link for LAN and SAN connectivity. If you know the Nexus family of data center switches, then you can start from a simple equation: UCS Manager software + Nexus hardware = Fabric Interconnect.
The counterpart of Fabric Interconnect in the blade server is the Cisco VIC (Virtual Interface Card) and IO module, which not only take care of connectivity, but also integrate the management of the relevant part of the solution. The chassis of the UCS therefore only serves for power and cooling, which means that it lasts a really long time.
Should UCS Manager, with its clear and fast HTML5 interface (or perhaps an API?) Be an insufficient tool for you, you can solve your demands for managing multiple data centers with UCS Central. Alternatively, you can also use UCS Director if you are building a cloud-like environment and require a very high degree of automation.
Yes, the physical server is what we pay the least attention to in Cisco's server solution, everything interesting has moved to the software. The great thing is that you can try UCS without having to buy it, just download the UCS Platform Emulator.
How from UCS to hyperconvergence (HCI)?
Hyperconvergence is usually defined by the integration of compute, network, storage, hypervisor parts of the data center and, above all, their management from one place. This definition is not entirely clear and is therefore interpreted differently.
Does the compute and management part include, for example, the firmware of individual server components, in other words, I would like a simple solution for rolling upgrades of all components? Does the network part include physical networking, not just software? Storage also includes configuring the access of a hyperconverged solution to my existing storage, so I want the solution to be integratable with the existing infrastructure and not create another separate force?
And what about manufacturer support? Does it make sense for me to have a single point for such a solution, where will I turn with all the problems, without managing cases and ensuring communication from more than one manufacturer? In my opinion, the answer to all four questions is unequivocally yes.
Hyperconverged solutions integrate the above to save us time and human resources. The administration department, albeit only a few parts, goes against this principle. Some manufacturers highlight a purely software solution - that is, you can use any element as the underlying hardware. If you go this route, the customer or partner is responsible for performance and compatibility. But how to ensure compatibility and the necessary performance? Ha! I will use validated solutions, but this narrows the term "arbitrary" somewhat. In addition, it does not solve all these problems by far, and so we get to a certain extent again to the disadvantages of converged data centers.
Of course, different approaches may suit different customers, and the single best probably doesn't exist. As you may have guessed, Cisco has taken a different path in meeting those who demand maximum simplicity, and the answer to these HCI requirements is Cisco Hyperflex.
UCS combines computer, networking, and management - Cisco is the only vendor to integrate physical networking into HCI. The same is true for Hyperflex, we only lack storage that is provided by HX data platform software, which is a distributed scale-out file system developed specifically for Hyperflex:
Out-of-box allows fast clones, instant creation and deletion of snapshots without affecting performance
Compression and deduplication on small blocks. This makes this file system highly space efficient, and if you require the maximum amount of compression, these calculations can be accelerated by FPGA cards.
Data overwriting uses available free space and is written when there is enough data to write to reduce the number of IO operations and increase performance.
No need to move data when migrating VMs, the file system is designed so that any node can read from any part of storage without losing performance
It uses the fastest possible access to writes, data is captured and distributed immediately, not after being written to the local node
Optional 2 × or 3 × replication factor
Automatic repair, tracking and rebalancing
HX Data Platform is therefore a modern file system, directly designed for hyperconvergence. In addition, it does not require the use of only converged nodes. If you do not need more space, but only computing power, only compute nodes can be extended.
No comments:
Post a Comment