Basic deployment of SAS Switch technology
Imagine that your company has hundreds of decisions, managers, and employees who need to exchange information and negotiate with each other. You may choose to arrange your company's office headquarters in a downtown location such as Shanghai Lujiazui. Although the land price is expensive, the convenient location allows the company to operate in a timely and efficient manner. I also envisage that your company has a large number of goods that need to be stored and transported. Most of the company's warehouses will choose to be located near the airport terminal in the outer suburbs. The land price is cheap and the transportation is convenient.
When designing computer systems, people also noticed the concept of land prices. Simply think that CPU, memory, and disk are all equivalent computer components. In fact, the land price is also in the computer system. Simply put, the closer to the CPU, the higher the speed, low cost, and low latency of the I/O connections that can be provided, so the physical space value close to the CPU is higher, and the space away from the CPU, the lower the value of physical space. In turn, the external storage of the data center can be compared to the data warehouse in the outer suburbs.
The concept of land prices in a typical blade server is very prominent. Typically, a blade server is the basis of a system with a multi-slot chassis with a high-speed interconnect backplane. Usually the chassis is expensive. Usually manufacturers will provide CPU blades (server nodes) with different configurations, storage blades (multi-bay JBOD), and switch blades for users to choose. There is no need to calculate the price of a specific product. We can simply make the following simple conclusions by considering the expensive price of the blade server:
1) It is better not to have a disk on the CPU blade. The same valuable space, more CPU and memory modules are usually more reasonable.
2) Storage blades are also best avoided, and the same valuable slots are not reserved for more valuable CPU/switch blades.
3) Externally connect a JBOD to provide the storage space required by all CPUs of the entire blade server, which is economical and easy to expand.
In the hardware systems of large data centers, the same land price factors are also working. The figure below shows that by reducing the number of disks per node, you can double the density of the server. The single node is accommodated from the 1U position and is upgraded to 1U2, which can be further upgraded to 1U3, 1U4. At present, the highest density in the industry can accommodate 160 server nodes per rack, a total of 320 CPUs, which is more than 4 times higher than the traditional rack density.
The improvement of efficiency, in addition to the more fully utilized physical space, also brings opportunities for device sharing within the appropriate package boundaries. For example, Facebook's 4 nodes can share the same Ethernet card, PMC+Intel RSA. In the reference design, several NVMe SSDs can be shared among the 4 server nodes. Because when these nodes are densely packed in a 1U space, it is possible to achieve high-speed interconnection of PCIe through inexpensive PCB connections to achieve device sharing without introducing expensive additional cable costs.
In the Scorpio 2.0 rack, the partner has implemented SAS Switch, JBOD and other specific products inserted into the rack, so it is very easy to set up a SAS Switch-based storage solution:
With such a physical architecture, how to optimize resource utilization efficiency in a rack? Specifically, how do you decide how many server nodes to configure before deploying a rack, and how many JBODs? To put it simply, the answer to this question lies in the resource matching characteristics of the specific application. The most important one is the ratio of the number of CPUs to disks. The ratio of this ratio can be calculated by evaluating the performance of the application to determine the initial deployment ratio of the device, and during the long-term operation after deployment, the dynamic operation state of each CPU is dynamically optimized.
The process is as follows:
The basic approach is to deploy SAS Switch technology in the enterprise and data center.
Pharmaceuticals,2-Methyl- Propanoic Acid Monohydrate Price,2-Methyl- Propanoic Acid Monohydrate Free Sample,Pure 2-Methyl- Propanoic Acid Monohydrate
Zhejiang Wild Wind Pharmaceutical Co., Ltd. , https://www.wild-windchem.com