Rebirth: Starting from Lighting Up the Tech Tree

Chapter 89 The Beam Dilemma

The next morning, Zuo Cheng went to the Sky Dome project's development lab. The lab was brightly lit; Tang Xu had clearly stayed up all night again.

Tang Xu's workstation was filled with printed simulation logs, with anomalies marked with a highlighter. When he saw Zuo Cheng arrive, he stood up somewhat embarrassed.

"Mr. Zuo, I've been investigating the beam switching delay issue for almost two weeks, but I still haven't found the root cause," Tang Xu said. "Our initial assessment is that during multi-satellite parallel processing, there's resource contention between the beam management module and other modules, causing switching commands to queue and wait, increasing the delay from 50 milliseconds to 120 milliseconds."

Zuo Cheng sat down and began reviewing the simulation logs. The logs were very detailed, recording the state every millisecond. Red indicated abnormal segments, green indicated normal segments, and yellow indicated critical segments. Zuo Cheng stared at the data for half an hour, his gaze locking onto a red area, before calling Tang Xu over.

"Look here," Zuo Cheng pointed to a section in the log, "After the beam switching command was issued, it entered the waiting queue at 37 milliseconds and wasn't executed until 107 milliseconds. What module was consuming resources during that 70-millisecond wait?"

Tang Xu leaned closer to look: "It's a spectrum sensing module. It's performing real-time spectrum scanning, consuming a lot of computing resources."

"Does spectrum sensing and beam management share a single computing unit?"

"Yes, under the current architecture, the two modules share GPU resources," Tang Xu said. "There was no problem when the number of satellites was small in the second phase, but with 480 satellites operating simultaneously in the third phase, the computational load for spectrum sensing is four times that of before, which fully utilizes the GPU resources."

Zuo Cheng immediately saw the crux of the problem. The resource contention wasn't an algorithmic issue, but rather a problem with the architecture design. If beam management and spectrum sensing shared the same GPU, beam management would be overwhelmed by the surge in computational demands for spectrum sensing.

"The two modules must be resource-isolated," Zuo Cheng said. "Beam management will be allocated to a separate set of GPUs, and spectrum sensing to a separate set. This way, they won't interfere with each other."

"But our hardware platform only has two GPUs," Tang Xu said. "One runs signal processing, and the other runs spectrum sensing and beam management. If we need to isolate them, we would need to add another GPU, which would increase the hardware cost by 15 percent."

A 15% increase in hardware costs translates to 18 million in a 120 million contract. That's a significant amount, but if the beam switching delay isn't resolved and acceptance testing fails, the losses will be even greater.

"Let's not worry about the cost for now; solving the problem is the priority," Zuo Cheng said. "I'll talk to Lanwan about the hardware aspects and see if we can apply for a change. You should design the resource isolation solution first to reduce latency."

Tang Xu nodded: "Understood. I'll change it today." After saying that, Tang Xu immediately returned to his workstation and began to redesign the architecture.

Zuo Cheng spent the entire morning in the lab discussing resource isolation solutions with Chen Hao and Tang Xu. The final solution was: signal processing would have exclusive use of GPU group A, spectrum sensing would have exclusive use of GPU group B, and beam management would have exclusive use of the newly added GPU group C. The three GPU groups would operate independently and without interference.

"What model of GPU do we need for the new GPU group?" Chen Hao asked.

"Like the existing Group B, the model is standardized for ease of maintenance," Zuo Cheng said. "I'll handle the hardware procurement; you guys focus on improving the software."

"Okay." Chen Hao and Tang Xu nodded at the same time.

Zuo Cheng returned to his office, closed the door, and opened the system panel. He carefully examined the detailed description of the beam management blades on the aerospace communications branch.

While the technology on the blades does include beam management, it provides optimal beam switching algorithms, not hardware architecture design. If the architecture is flawed, technology can't help.

He closed the system panel and called Lu Mingyuan.

"Mr. Lu, there's a technical issue regarding the third phase of the Sky Dome project that I need to discuss with you. The beam management module experienced excessive latency during multi-satellite parallel testing due to GPU resource contention. We plan to add a set of GPUs for resource isolation, which will increase the hardware cost by approximately 1800 million."

Lu Mingyuan was silent for a few seconds: "How much is the delay excessive?"

"The increase from 50 milliseconds to 120 milliseconds is a veto item for acceptance testing."

"This must be resolved," Lu Mingyuan said. "I've discussed the increased costs with General Manager Zhou; it shouldn't be a problem. Submit the technical solution as soon as possible, and I'll arrange for its review."

"Okay, thank you, Mr. Lu."

After hanging up the phone, Zuo Cheng breathed a sigh of relief. Blue Bay's positive attitude showed they still trusted 402's technical capabilities. The impressive 94-point score in the second phase of Sky Dome was there for all to see; a single architectural adjustment wouldn't shake Blue Bay's confidence. As long as the solution was reasonable, increased costs weren't a problem.

But Zuo Cheng knew in his heart that this beam switching problem exposed a bigger hidden danger: the hardware architecture design of 402 was not mature enough. The technology tree had given him the most advanced algorithm, but the engineering implementation still depended on the team. If the team's architecture design capabilities could not keep up with the speed of algorithm evolution, similar problems would arise again in the future.

He needs to improve his system architecture skills. Perhaps, during the next technology radar scan, he can pay special attention to whether there are any talents and technologies in system architecture.

Just as he was thinking, Han Lu pushed the door open and came in.

"Mr. Zuo, the due diligence team from Ginkgo Capital will arrive tomorrow and will need your cooperation for a technical interview," Han Lu said. "Also, Blue Bay Industrial Fund has replied; they are very interested in strategic investment but hope to see the latest progress on the third phase of Sky Dome before making a decision. However, I think it should all be fine in the end."

"The beamforming issue with Sky Dome is still being resolved, so we can't show it to them yet," Zuo Cheng said. "Arrange the technical interview with Ginkgo Capital first, and we'll talk to Blue Bay after the beamforming issue is resolved."

"Understood." Han Lu left.

Zuo Cheng looked at the system panel and made a decision. Tomorrow, Ginkgo Capital will be conducting due diligence; an investment manager and technical advisor will be visiting. He can use his technology radar to scan for any worthwhile technical capabilities to learn.

The technology radar scan costs 5 points, and if there's a technology worth replicating, it will cost additional points. But he currently has 145 points, which is enough.

He shut down the system panel and began preparing the materials for the technical interview he would need tomorrow. Fundraising, the Sky Dome, and smart cities—three projects were progressing simultaneously, and none of them could afford to go wrong.

This is the daily life of an entrepreneur: always walking a tightrope, never able to stop. But as long as the technology tree is still there, they have the confidence and strength to keep going.