This article was originally written by IPFS Force Zone Internet Data Center ( IDC ) refers to a service platform with complete equipment, professional management and complete applications. The distributed storage network built by Filecoin also requires professional venues to house equipment. So what is the difference between Filecoin 's IDC and traditional IDC , and what matters need to be paid more attention to? The following is a sharing by Tiangui, a network architect of IPFS Force Zone, on the topic of "Data Center Planning in IPFS ( Filecoin ) Application Scenarios". First, Tiangui divided the weights of various factors that affect Filecoin mining. He said that the factors that currently affect Filecoin mining can be summarized into five aspects, including: mining farm IDC , IDC network, mining machine hardware configuration, operation and maintenance capabilities, and mining software. Among them, the weight of mining farm IDC and IDC network accounted for 15% each, which was the main content of this sharing. The site selection of traditional IDCs is very particular. They must be located in places with suitable climates and away from places prone to natural disasters such as rivers, typhoons, earthquakes, etc. There are two main types of IDC providers in the market: 1. The operator's computer room is aimed at large customers and has resource advantages, but the service is simplified, only providing simple services such as wiring, racking, and restarting. If you need higher quality services, you need your own company to send professional operation and maintenance engineers; 2. The third-party IDC computer room has a professional operation and maintenance team and standardized collaboration processes, which can provide high-quality services. IDCs generally have stable and reliable network and power resources to ensure the stable operation of data centers. However, Filecoin has multiple roles at the bottom layer, and all clusters are distributed and parallel. This requires more work to be done on the network architecture level for Filecoin mining to achieve the requirements of no blocking, no packet loss, and low latency. The idea of network construction is to design different network architectures according to the scale of the mining pool. For example, the two-layer architecture based on DRNI is suitable for the scale of 300-500 servers; the flat structure of the three-layer routing is suitable for the scale of 1,000 servers, with the characteristics of simple structure and convenient later maintenance; the 10G access model based on POD is suitable for the scale of more than 30,000 servers, with the characteristics of very strong network expansion capabilities; the last one is the 25G access model based on POD , which is also suitable for the scale of more than 30,000 servers, but is mainly for cloud computing and big data platforms. After choosing from the above four main network structures, the next step is to design the detailed network architecture. In the Internet egress area, two lines are configured with static routing to connect to the IDC network, and firewalls are deployed for security protection; in the core switching area, highly reliable frame switches are used, which support multi-power supply access, large cache and VOQ mechanism; in the service access layer, according to the role of the cluster, the convergence ratio of the computing cluster 4*40G uplink and the acceleration cluster 4*40G uplink can be controlled at 1:3 , while the convergence ratio of the storage cluster 4*100G uplink is 1:1 . The main principle of the convergence ratio is that the greater the bandwidth requirement, the lower the convergence ratio must be. In the process of Filecoin mining, there will be computing tasks, and one computing task will be split into multiple subtasks for parallel computing. The "many-to -1 " scenario, that is, multiple tasks access a port at the same time, will cause congestion, which will lead to packet loss. The solution to this problem is to enable the core switch to support port caching, solve the congestion problem through the queue scheduling mechanism, and ensure 100% packet forwarding. Finally, Tiangui also popularized basic knowledge about IP network address settings, air duct environment, equipment placement, etc. The main purpose is to achieve refined management and improve scalability through standardization. Statement: This article is an original article from IPFS Force District. The copyright belongs to IPFS Force District. It may not be reproduced without authorization. Violators will be held accountable according to law.
Tip: Investment is risky, so be cautious when entering the market. This article is not intended as investment and financial advice. |
