Filecoin mining: Everything is ready except the mainnet

Filecoin is very popular.

Just like stacking hard disks to provide storage capacity, Filecoin stacks countless resources in the industry, and practitioners are waiting for the resources to be realized. Even though Filecoin has had many believers for a long time, many people have left, but there are still more people entering.

The current state can be clearly expressed in eight words: everything is ready, only the mainnet is missing. Investors and companies that have invested real money have already built a world.
What people in the industry must think about

It has to be said that this "world" is built by mining machines. Storage mining machines build storage networks, which is a dream about future commercial infrastructure.

The Filecoin network is based on the IPFS protocol, which gathers the computing and storage capabilities of mining machines to build a network. In order for miners to work in an orderly manner, the network needs an incentive method, that is, to "evaluate" what miners do and prove it, and then reward miners. This is why proof of replication and proof of space and time appear. When miners actually store data, these proofs are written into the block, and they receive "reasonable and fair" incentives.

The hardware parts of the mining machines, miners and network are ready.

Since December 2019, the Filecoin testnet has gone through two testing processes. In the first phase, many testnet nodes competed to be on the list, striving to prove that "their technology is strong enough." After the second phase restart, the nodes calmed down.

Due to the algorithm design of the Filecoin network, nodes need to achieve effective storage. The first requirement is that the stored data is valid data, and the second requirement is that the data must be packaged, stored, and reported to the main network quickly and successfully. The data must then be saved and kept online at all times. This is a challenge for nodes from technology to service, and it is also where nodes need to compete continuously.

Before the mainnet was launched, the official announced 4 million FIL rewards for miners and the five continents plan. From a competitive perspective, there will inevitably be a surge in effective computing power before and after the mainnet is launched. Zhu Gong, the head of the 95 mining farm, told Golden Finance that his mining farm now has a computing power of about 50PB. Before the mainnet is launched, he will reserve the existing computing power to 200PB. Hu Feng, COO of Spacetime Cloud, predicted that after the mainnet is launched, the effective computing power will soon rise to more than 1,000PB, and the participating computing power of the entire network may reach 10,000PB.

1475 co-founder Andy sees this slightly differently.

"According to the evaluation at the end of June, the machine access capacity of the entire market can reach about 100PB in 20 days. In terms of capacity, there may not be much change when the main network is first launched, because all the machines that can be put into use have been put into use. Therefore, the main network launch is at most 1.2~1.5 times higher than this time (second test). If there is no problem with the supply chain and the main network is relatively stable after it is launched, the daily effective storage growth will be about 15~20PB."

Everything is ready, except the mainnet. Only when the mainnet is online can the incentive model and rewards be realized. Every node is analyzing, judging and choosing, and the market is progressing rapidly, which is due to competition.
The significance of testing

For a network that has not yet been launched, testing the network is an inevitable process and is crucial to the future development of the network. Hu Feng, COO of Spacetime Cloud, explained, "In fact, the significance of testing is to test the security and stability of the network, and to verify the hardware configuration, software performance, and cluster solutions."

Hu Feng also revealed that the current testing cost is not as high as everyone imagines. The main cost of testing is the rental of the IDC computer room. Generally speaking, the monthly rental of a cabinet is around 4,000-5,000 yuan. If the equipment in 10 cabinets is used, the cost of testing is about 50,000 yuan.

Therefore, during the testing phase, most companies chose to reserve sufficient computing power. For example, Hou Feng, the founder of Maike Storage, said that the computer rooms that Maike Storage has built for Filecoin mining are not open.

In the technical testing process of the nodes, the core part mainly tests the task grabbing and data packaging process. Task grabbing is actually a process of competing for computing power, "this part is similar to PoW mining," Hu Feng added.

Regarding the data encapsulation part, Andy, co-founder of 1475 mining pool, explained the technical breakthrough part to the Golden Finance reporter.

"We split and rebuilt the underlying code, and split it into very fine parts. For example, for stages like "P1" and "P2", we also split them more finely. At most, there are more than 60 optimization points in total, and each optimization point can improve efficiency by 5%-20%." In addition, Andy also revealed that the P2 stage will be GPU-based, and the zero-knowledge proof part will be outsourced to reduce resource consumption.

Ronglai Technology, which is also a test node, has roughly the same idea. Ronglai Technology CTO Heige said: "On a superficial level, it is the hardware configuration and unique cluster architecture, and on a deeper level, it is the running software or core algorithm. We have changed more than 20% of the official source code."

In the market, we often hear descriptions of packaging speeds, and many nodes have announced packaging data: the fastest is 2 hours or 1 hour. However, it seems that there are misunderstandings in the packaging process.

As Andy believes, there are some suggestions in the market that the packaging process can be completed in 2 hours, but this does not mean that it is a solution that maximizes commercial value, because although it can be achieved through overclocking, it is not a long-term solution. "The high speed of a single thread does not mean that you will have the highest output in the end, because there are many tasks, and you must try to parallelize them as much as possible. The fastest single process does not mean the optimal solution."
Join the game one after another, each showing their strengths

Facing the future of Filecoin, every node that strives to be the first or a big miner has to invest a lot. In our traditional mindset, as a network node, you may only need a cloud server. For example, in the EOS node election that year, all nodes used cloud servers because it was the most convenient and could easily earn profits. The same is true for Bitcoin nodes. However, Filecoin has upgraded the requirements for nodes.

The first is hardware. The storage space requirement is at the PB level. Large miners require at least 5PB. In view of the cost, hard disks are used to build the hardware. However, hard disks alone are not enough. Computing power is also required, that is, adding GPUs to cooperate with hard disk arrays. In order to better store and read data, stable bandwidth and equipment management are also required, which requires a professional computer room to achieve.

This is what Filecoin requires of nodes. Investors need to consider long-term investment and output.

A node needs to prepare a large amount of computing power to meet the needs of the main network, which means that the node needs to assemble mining machines and build its own IDC computer room, and these hardware equipment can be turned into cloud computing power or mining pool business. Each process can be closely connected.

Just like Maike Storage, because it targeted the server supplier market, built a factory to produce mining machines, and then built its own professional IDC computer room. Seeing the market opportunity of Filecoin, Maike Storage founder Hou Feng led the team to quickly enter the Filecoin track.

From the perspective of corporate structure, the corporate structure design of Maike Storage is very similar to that of Bitmain. The hardware business is carried out as usual, while the cryptocurrency Filecoin is both a business and an investment. Hou Feng told the Golden Finance reporter that Maike Storage is aiming for the listing goal, so the current business model is to develop multiple processes simultaneously.

He expects that Filecoin will develop a separate mining ecosystem like Bitcoin, and may even surpass Bitcoin. This will rely on more nodes to integrate mining machines, mining pools and other businesses. This is also the basis for the development of the network's application ecosystem in the future.

For example, the future plans of the 1475 team also include plans to incubate developer projects.

Andy said that the focus is still on the mining ecosystem, and application plans need to wait until the market matures before being incubated in the form of funds.
Performance and cost

Filecoin's mining ecosystem is ready. For every participant or investor, the most important thing is still the cost, and the cost is something that needs to be repeatedly considered and studied.

Just like Andy thinks, calculating the cost per terabyte may be a "pseudo-concept", because from a technical point of view, a machine can be connected to many hard disks, and the more hard disks are piled up, the lower the cost per terabyte. Therefore, to judge the cost, we still have to look at the overall solution, rather than just the storage hard disk capacity, so "how many coins are produced per terabyte" seems to be a false proposition.

The importance of cost issues is the same for enterprises. Hou Feng, founder of Maike Storage, said that although the current performance depends entirely on the hardware stack, it is also mixed with the issue of cost performance. "For example, 8-core and 12-core, 12-core may be faster than 8-core, but its energy consumption is high. If the cost performance is not high, there is no need to replace 8-core with 12-core."

In addition, Hou Feng also revealed, "We may use some supercomputing servers for single-purpose, and the rest can be handed over to other machines for storage services, and they are matched together. It's like using a Rolls-Royce to rob, but I use a Mercedes-Benz to actually run."

Different roles and perspectives face different costs.

Mr. Zhu, the person in charge of the 95 mining farm, was still mining Bitcoin and Ethereum in 2018. At the end of 2018, he met Tianru Technology. After discovering that Tianru Technology had advanced technology in Filecoin, he joined the company with his mining farm and capital.

Mr. Zhu has been focusing on mining farm operation and maintenance for four consecutive years. He has entered the storage mining circle and has a thorough understanding of the business of Filecoin mining farms and mining machines. He knows the equipment and costs like the back of his hand.

It is understood that for the four officially recommended CPUs, Zhu's team tested two models, one is 3950X and the other is 3970X. 3950X can be equipped with 128G memory, and 3970X can be equipped with 256G memory. Overall, 3970X performs better.

"These two are desktop CPUs, and there are also server-level ones, mainly the 7474 and 7402 series." The comparison results show that the server is indeed faster than the desktop, because the 3970X and 3950X are easy to overclock, but the server does not allow overclocking. The server can also support larger memory (512G or even 1T, 2T).

It is understood that the current computing power of the 95 mining farm is about 50PB, and it will be expanded to 200PB when the main network is launched around August. Mr. Zhu said that there will be "greater development" 3 or 6 months after the main network is launched, but he is very sure about this goal at present.

According to current market cost calculations, it is roughly estimated that each mining machine is about 100T, and 200PB of computing power requires about 2,000 100T mining machines. Based on the cost price of at least 50,000 yuan for a 100T mining machine, in the computing power reserve plan of 95 mining farms, the mining machine part alone will require hundreds of millions of funds to complete.

In addition to the basic hardware, the cost also needs to be considered in terms of operation and maintenance, but this part is also a matter of opinion.

Hou Feng, founder of Maike Storage, believes that Filecoin is far from "considering electricity costs" in its early stages. "I think based on the current consensus, there is basically no need to consider operating costs. If Filecoin can be applied, it will be revolutionary. It will be a scenario at the level of Bitcoin, or even beyond the level of Bitcoin."

Zhu Gong of the 95 mining farm is used to standardizing the operation and maintenance of the Filecoin mining farm. "The mining farm looks simple but it is full of pitfalls."

Zhu Gong pointed out that the first focus of a mining farm is to build a network structure, ensure the I/O speed of each machine, test the speed of each gateway interface, perform read tests on the inside of each machine, and monitor the real-time status of each hard disk. "Because if any abnormality is found in the sector during the mining process, we must deal with it immediately."

Therefore, when doing mine operation and maintenance, managers hope that the machine configuration is unified and the on-site manager is used for automatic inspection, which can greatly reduce manpower. In addition, there are three major energy consumptions in the computer room: electricity, bandwidth, and labor. In order to reduce consumption, some internal data exchange and internal transportation may be performed to reduce bandwidth consumption and reduce power consumption.

However, most mining farms are already extremely standardized. For example, the computer room of the 95 mining farm is on the first-level national backbone network of the Wuxi National Sensor Technology Park. The computer room of Maike Storage can reach the T3 level and provide a very cost-effective T2 level computer room.