What are the advantages of Hyperconverged Infrastructure (HCI) compared with traditional infrastructure?

I. Comparison of Architecture and Resource Management Mode

Taking SmartX hyper-converged products as an example, the differences in deployment and resource management modes between the lower hyper-converged architecture and the traditional architecture are given respectively.

Figure 1 deployment differences between traditional architecture and hyper-converged architecture Figure 2 differences in deployment and resource management modes between traditional architecture and hyper-converged architecture

Compared with the traditional FC SAN architecture, hyper-converged architecture has the following significant changes:

1. Use standard and easy-to-maintain x86 servers and 10 Gigabit Ethernet switches instead of proprietary storage hardware and networks;

2. The core is distributed storage, and each server is a storage controller. It should be noted that each node needs to configure SSD disk as cache, and the cache capacity is much larger than the volatile memory of traditional storage, and the concurrency of multiple nodes brings stronger aggregation performance;

3. Logically, storage is no longer a different storage space composed of RAID, but a unified and flexible storage pool with stronger expansibility;

4. Another benefit brought by distributed storage is that performance and capacity can be configured on demand in a unified resource pool, and data can be automatically balanced;

5. Compute virtualization and storage are deployed on the same server node.

Second, distributed storage+virtualization deployment is the essence of hyper-converged architecture.

For hyper-converged architecture, because of its unique name, many users have a misunderstanding: the core of hyper-converged architecture lies in fusion, so the more functions of fusion, the better. However, through the comparison of the above architectures, we can see that the transformation of hyper-converged architecture is firstly the substitution of distributed storage for traditional storage, and other advantages brought by this substitution (such as building based on x86 server, concurrency, easy expansion, etc.).

Of course, the unique deployment mode of distributed storage and virtualization further simplifies users' IT architecture, reduces the use cost and the difficulty of operation and maintenance, and these values also greatly accelerate users' acceptance of distributed storage mode.

3. What improvements will hyper-converged architecture bring to IT infrastructure? Why?

The local upgrade and hyper-converged architecture of different products have brought comprehensive improvement to IT infrastructure, which will be explained one by one below.

1, which effectively improves the reliability.

A common misunderstanding of software-defined data center architecture by many enterprise customers: Is the system built by x86 server reliable? Although the x86 server itself has a single point of failure, hyper-convergence itself works in cluster mode. The first problem to be solved in its core distributed storage system is to build a large-scale system with higher reliability by using technologies such as multiple copies, which is the core of building a software-defined data center.

Moreover, compared with traditional centralized storage, distributed storage has many advantages in reliability, and the comparison is as follows:

In addition, SmartX, Nutanix, VMware and other vendors provide enterprise-class storage services ranging from data checksum to rack awareness, active and passive, remote backup and other functions, as well as enterprise-class high availability and data protection solutions integrated with third-party products.

2. Concurrency performance is greatly improved and IO delay is reduced.

The following still introduces why hyper-converged architecture can improve the performance of SMTX operating system and reduce access delay.

Among them, the distributed architecture improves the aggregation performance of the whole system; SSD cache mechanism improves the access performance of single node; I/O localization mechanism has advantages only in hyper-converged deployment mode, which can further reduce access delay. It should be noted that I/O localization is mainly supported by SmartX and Nutanix, but hyper-convergence based on open source products such as Ceph cannot be supported.

To give a concrete practical example, due to the bottleneck of dual control, HP 3PAR 8440 has basically reached the maximum performance among eight SSDs, but the reading and writing performance can be achieved only by using four SmartX hyper-converged nodes.

3. The expansibility has been greatly improved.

As mentioned above, the core distributed storage of hyper-converged architecture greatly improves the scalability compared with traditional storage, including the following features:

Among them, the support of heterogeneous nodes needs to be confirmed with vendors, similar to SmartX, but many vendors do not provide such support.

4. The difficulty of operation and maintenance is greatly reduced.

In view of the difficulty of operation and maintenance, we can compare the two architectures for the whole operation and maintenance cycle:

As can be seen from the above figure, hyper-converged architecture not only automates a lot of operations in the whole product operation and maintenance cycle, but also has short time and high efficiency. IT can effectively reduce the requirements for personnel, liberate IT personnel and enable them to carry out more innovative activities.

5. Reduce procurement costs and total cost of ownership.

In terms of the cost that customers are most concerned about, the procurement cost of server+hyper-converged software (or hyper-converged all-in-one machine) has been greatly reduced compared with that of server plus traditional high-end storage. But in addition to procurement costs, hyper-convergence has more advantages in total cost of ownership.