owlsbooks Chapter 211 High-Performance Web Servers
Chapter 211 High-Performance Web Servers
Just as Ballmer was doing everything he could to stop the Star System from expanding its market share, Ling Yun was completely unaware that his opponent had begun to lose patience and was resorting to some underhanded tactics to deal with him.
Even if Ling Yun knew, he wouldn't interrupt his plan; he would only say dismissively, "Is that all?"
At 9:20 a.m., the conference room on the third floor of Xingchen Technology was already full of people.
On either side of the long table, on the left, are the management team: CEO Ling Yun, Operating System Director Eric, Browser Director David, Ecosystem Tools Director Carly, Finance Director Emily, and Administration Director Mark.
On the right are the board representatives: Michael Dell of Dell (participating by phone), Richardson of Goldman Sachs, Ben Rosen of Compaq (via video link), Jerry Sanders of AMD (by phone), and Louis Platt of HP (via video).
Each person had a three-page meeting document in front of them, with the cover titled: "Proposal for a High-Performance Web Server Project".
Ling Yun sat at the end of the long table, not wearing a suit, but a dark gray sweater with the sleeves rolled up to his elbows. In front of him, besides the materials, was an open ThinkPad with charts displayed on its screen.
"Everyone's here," said Mark, the administrative manager, glancing at the clock on the wall. "Shall we begin?"
Ling Yun nodded.
Mark pressed the button on the recorder, and the red indicator light illuminated. "Star Technology Third Board Meeting, December 11, 1997, 9:21 AM. Attendees…"
After reading the list, he looked at Ling Yun.
"There's only one core topic today," Ling Yun said, his voice low, but the meeting room immediately fell silent. "The new project proposal. Everyone has seen the materials; I'll just briefly go over the core content."
He turned to the whiteboard behind him, picked up a marker, and drew a simple layered diagram:
Application layer (games/instant messaging/e-commerce)
↓
Web server
↓
Operating System (Starry Sky)
"The current infrastructure of internet services," Lingyun pointed to the middle layer with the tip of his pen, "mainly relies on Apache. It's good, open source, and stable, but it has problems."
He wrote down two phrases: "low concurrency" and "high resource consumption".
"Apache uses one process or thread per connection," Lingyun turned around, "and when concurrency increases, memory and CPU can't handle it. Our data shows that for even slightly active websites, the response speed starts to drop when there are more than a thousand concurrent users. If it exceeds five thousand, it basically crashes."
David raised his hand: "But most websites don't reach that level."
"We can't reach that now," Lingyun said. "What about next year? The year after? The number of internet users doubles every six months. Moreover, some scenarios inherently have high concurrency requirements."
He walked back to his seat and tapped a few times on his ThinkPad. The projection screen at the front of the conference room lit up, displaying a table:
Existing solutions for scenarios with concurrent requirements
Instant messaging: 5+ (but not enough)
Online game lobby with 2+ players but less
E-commerce promotions exceeding 10 users cause system crash.
Large communities with 3+ residents are developing slowly.
The numbers are in red.
"StarTalk Instant Messaging," Lingyun said, "is our own product. We currently have 500,000 users, with a peak concurrent user count of 12,000. Apache is already struggling. What if we reach 5 million users and a peak concurrent user count of 100,000 next year?"
No one spoke.
"Regarding StarCraft's Battle.net functionality," Lingyun continued, "Blizzard reported that their matchmaking server, built using Apache, experienced significant latency during peak hours. If we want to develop a large-scale multiplayer online game, we must resolve this bottleneck."
Goldman Sachs' Richardson adjusted his glasses: "How big is the market?"
Lingyun switched to a slideshow. A bar chart showed the growth curve of the number of websites worldwide: 20,000 in 1995, 250,000 in 1996, and an estimated 1.5 million in 1997. Small print next to it indicated the data source: Netcraft.
"The number of websites is growing exponentially," Lingyun said. "Ten percent of them will become active sites, and five percent will need high-performance servers. Right now, this market is empty—there are no mature products. Apache is the only option, but it's not enough."
"What about our competitors?" Compaq's Ben Rosen asked in the video. "What about Microsoft?"
"Microsoft has IIS," Lingyun said, "but it only supports Windows NT. Moreover, it's designed for enterprise intranets, not high-concurrency internet scenarios. More importantly, IIS is not open source, has poor customization capabilities, and is difficult to extend."
He paused, then looked at everyone: "Our opportunity lies in creating an open-source, lightweight web server optimized for high concurrency. It will run on the Starry Sky system, but can also be ported to other systems."
AMD's Sanders' voice came through the phone: "How technically challenging is it? What are the difficulties?"
"Architecture," Lingyun said, "Apache uses a multi-process model, one process per connection. We need to use an event-driven model, asynchronous non-blocking I/O, to handle a large number of connections with a small number of threads."
He drew a simple diagram on the whiteboard:
Master process
— Worker Process 1 → Processing connections A, B, C...
— Worker Process 2 → Handling connections D, E, F...
— Work process N (Worker)
"The worker processes are independent," Lingyun explained. "If one crashes, it doesn't affect the others. Memory can be shared, reducing overhead. Hot updates are configured so that services don't need to be restarted."
Eric raised his hand: "Does the kernel need to be modified?"
"Some optimizations are needed," Lingyun said, "mainly the network stack and event notification mechanism. Your team needs to cooperate."
"How long will the development take?" Michael Dell asked.
"We'll release the first usable version in three months," Lingyun said. "We'll reach production environment standards within six months. Our initial goals are: support for 50,000 concurrent connections, 70% lower memory usage than Apache, simple configuration, and support for reverse proxy and load balancing."
Finance manager Emily turned to the third page of the documents: "Boss, what's the budget?"
"The first phase investment is $1.5 million," Ling Yun said. "The main cost is human resources. We'll build a team of 20 people over six months. Additional investment will be made later based on progress."
"Are there any predictive models that yield returns?" Richardson asked.
Lingyun switched to the last slide. A simple financial model:
- Direct revenue (enterprise license, technical support): Projected cumulative revenue of US$50 million over three years.
- Indirect revenue (driving sales of the Star System, enhancing the ecosystem): difficult to quantify, but "significant".
- Strategic value: Filling critical infrastructure gaps and establishing technological barriers
"More importantly," Lingyun added, "if we have the best web servers, all internet companies that need high performance will consider the Star System, which is a key link in the closed-loop ecosystem."
The meeting room fell silent.
People looked down at the materials or stared at the slides. Some people were writing or drawing on paper.
Carly was the first to raise her hand: "I support it. Xingyu's backend is currently limited by Apache. If we had our own servers, product iteration would be much faster."
David nodded: "Browser and server optimization can enable more features, such as persistent connection push."
Eric said, "It's technically feasible, and the kernel team can provide support."
Finance officer Emily tapped a few keys on the calculator, then looked up: "The budget is reasonable. But I suggest allocating funds in phases and setting milestones."
On the board's end, Ben Rosen in the video window looked at the person next to him, whispered a few words, and then turned back: "Compaq agrees that our server business needs this kind of software."
AMD's Sanders: "Technically, AMD can assist with CPU-level optimizations, I agree."
HP's Pratt: "HP needs to assess the impact on its existing product line, but supports it in principle."
Goldman Sachs' Richardson put down the documents: "The market logic holds true, Goldman Sachs agrees."
Finally, it was Michael Dell from Dell. His voice came through the phone: "Dell agrees and recommends starting as soon as possible."
Ling Yun looked at Mark: "Let's vote."
Mark stood up: "Regarding the proposal for the high-performance web server project, please have the board members vote. Those in favor, please raise your hands."
In the video window, Ben Rosen raises his hand. Louis Platt raises his hand.
On the phone, Michael Dell said, "Dell agrees."
Jerry Sanders: "AMD agrees."
Richardson raised his hand.
Mark looked towards the management side. Carly raised her hand, David raised his hand, Eric raised his hand, Emily raised her hand, and Mark himself raised his hand.
Finally, Ling Yun raised his hand.
"Unanimously approved," Mark said, writing the result in the meeting minutes.