Chapter 25: The First Line
San Diego was blinding.
Not metaphorically—literally blinding. After a year of Seoul’s gray winters and filtered spring light, the Southern California sun hit Dojun like a flashbang the moment he stepped off the airport shuttle. The sky was a shade of blue that Seoul never achieved, the air was warm and dry and scented with something that might have been jasmine, and the convention center where ISCA 2007 was being held looked like a glass-and-steel monument to the future of computing.
Which, in a way, it was. Because inside that building, in four days, Dojun would present a paper that he had conceived in another life, proven in this one, and was now offering to the most critical audience in computer architecture.
He had arrived two days early, partly to recover from jet lag and partly because Kim Taesik had insisted: “Arrive early. Walk the venue. Attend the first day’s sessions. Let the environment become familiar before you perform in it.”
The hotel was a Holiday Inn three blocks from the convention center—the cheapest option in the area, booked by the university’s travel office with the budget-conscious efficiency of a public institution. The room was clean, functional, and had a view of a parking lot, which Dojun found oddly comforting. Parking lots were the same everywhere in the world.
He set his bag on the bed, pulled out his laptop—a proper laptop now, a ThinkPad that Bridge’s budget had justified—and opened the presentation for the forty-seventh time. Twelve slides. Thirty minutes including Q&A. The most important thirty minutes of his academic career.
His phone buzzed. A flurry of messages, accumulated during the flight.
Hana: Did you land? Is San Diego as beautiful as the movies? Send photos. I want to be jealous in real time.
Minjae: Bridge just hit 4,000 users. We’re managing fine without you. Hyunwoo hasn’t broken anything. Yet. Go destroy ISCA.
Seokho: Remember: “I don’t know” with a pause and a ceiling look. Practice in the hotel mirror. Also, the SIGCOMM reviewers sent preliminary comments. Positive. Details when you’re back.
Kim Taesik: Park. Present the work. Not yourself. The work speaks. You just need to hold the microphone. — Kim T.
His mother: Dojun-ah did you eat on the airplane? American airplane food is probably worse than university cafeteria. I packed kimchi in your suitcase. The small container, in the ziplock bag inside the plastic bag inside the paper bag. Don’t tell customs.
He checked his suitcase. There, nestled between his dress shirt and his good pants, was a small container wrapped in three layers of bags, exactly as described. He opened it. The kimchi smelled like home—garlic, red pepper, the deep fermented tang that no restaurant in the world could replicate because it came from his mother’s hands and thirty years of practice.
He ate a piece, standing in a Holiday Inn room in San Diego, twelve thousand kilometers from Namdaemun Market, and felt the distance between his two lives collapse into a single point.
I’m here, Mom. I made it.
The ISCA sessions were extraordinary.
Dojun attended the first two days as an observer, sitting in the back of ballrooms that held three hundred of the world’s leading computer architects. The presentations were dense, technical, and delivered with the particular confidence of people who had spent decades at the frontier of processor design. Intel researchers presented next-generation cache architectures. AMD engineers discussed multi-core optimization. Academic teams from MIT, Stanford, and ETH Zurich showcased work that would shape the industry for the next decade.
And Dojun, sitting in the back row, understood all of it. Every presentation. Every technique. Every advance. Because he had lived through the consequences of this research—he had built products on top of it, had hired teams to implement it, had seen it succeed and fail and evolve over forty years.
But understanding and contributing were different things. His paper was about branch prediction—a specific, well-defined subfield. The audience would include people who had literally invented the techniques he was critiquing. John Hennessy himself was a scheduled keynote speaker, though not present at the session where Dojun would present.
Small mercies, Dojun thought.
On the third day—his presentation day—he woke at 5 AM, showered, put on his good shirt and the tie that Hana had chosen for him (“Navy blue. Professional but not boring. If you wear the gray one, I’ll redesign it remotely via telepathy.”), and reviewed his slides one final time.
The session was at 2 PM in Ballroom C. He arrived at 1:30, found the podium, tested the projector, and stood in the empty room watching it fill.
By 1:55, every seat was taken. Word had spread—partly from the paper’s abstract, which had been flagged by several reviewers as “must-attend,” and partly from the simple novelty of a twenty-year-old undergraduate presenting at ISCA. The audience was overwhelmingly male, overwhelmingly senior, and overwhelmingly skeptical. They had come to be impressed or to find flaws. There was no in-between at ISCA.
The session chair introduced him: “Our next paper is ‘Dual-Mode Branch Prediction for Bimodal Workloads,’ by Park Dojun from Seoul National University. Park is an undergraduate researcher in Professor Kim Taesik’s group, and this represents, if I’m not mistaken, the first ISCA paper by a Korean undergraduate author.”
A murmur. Not hostile—curious. The kind of murmur that meant the audience was calibrating its expectations.
Dojun stepped to the podium. Three hundred faces looked back at him. The projector threw his first slide onto the screen—Hana’s clean design, navy blue on white, the bimodal distribution graph centered like a thesis statement.
“Good afternoon. My name is Park Dojun, and I’d like to talk to you about a flaw in how we think about branch prediction.”
He heard his voice through the microphone—steady, clear, younger than every other voice that had used this podium. And he began.
The presentation flowed the way good presentations do—not like a recitation but like a conversation, each slide building naturally on the last, each insight presented not as a revelation but as an inevitability. The bimodal distribution. The dual-mode predictor. The power savings. The simulation results.
When he showed the 7.2% accuracy deviation in the gcc benchmark, a professor in the front row—a gray-haired woman from Stanford whose work Dojun had studied in his previous life—leaned forward and made a note.
When he described the startup profiling mechanism for branch classification, two Intel engineers in the third row exchanged a look that Dojun recognized: the look of people who were already thinking about implementation.
When he showed the power consumption graph—15% reduction with 2.1% accuracy improvement—the room was quiet in the way that rooms are quiet when people are processing something they didn’t expect.
He finished in twenty-two minutes. Eight minutes for Q&A.
The first question came from the Stanford professor. “Your startup profiling approach assumes a warm-up period of approximately one thousand branches. For short-lived processes—shell commands, scripts—the profiling cost may exceed the benefit. Have you characterized this crossover point?”
“The crossover occurs at approximately ten thousand executed branches, which corresponds to roughly 0.1 seconds of execution time on current hardware. Processes shorter than that default to the standard single-mode predictor. The threshold is configurable.”
She nodded. Made another note.
The second question, from an AMD researcher: “How does the dual-mode predictor interact with speculative execution? If a branch is misclassified as regular when it’s actually irregular, the penalty could be worse than the standard predictor.”
“Misclassification has a bounded cost because the irregular-mode predictor is designed to conserve resources rather than optimize accuracy. A misclassified irregular branch in regular mode wastes some power but doesn’t cause pipeline stalls. A misclassified regular branch in irregular mode loses some accuracy but maintains pipeline flow. The asymmetry of the cost function is key to the design’s robustness.”
The AMD researcher typed something on his laptop. Rapidly.
Three more questions followed—each technical, each probing a different aspect of the design. Dojun answered them all. And twice—twice—he used Seokho’s advice.
“That’s an interesting question. I haven’t fully explored that direction yet.” Pause. Ceiling look. “But I suspect the answer relates to the interaction between branch density and pipeline depth. I’d need to run additional simulations to be certain.”
The audience accepted this gracefully. “I don’t know” from a twenty-year-old undergraduate was not weakness—it was maturity. It was the acknowledgment that research was a conversation, not a monologue.
The session chair called time. The applause was genuine—not thunderous, but sustained, the kind of applause that researchers gave when they recognized quality work.
Dojun stepped off the podium. His hands were shaking. Not from fear—from the particular adrenaline of having done something well in front of people who could tell the difference.
The Stanford professor approached him as he gathered his things. “Park. Your dual-mode design is elegant. I’m running a research group on power-aware architectures at Stanford. If you’re ever considering graduate school abroad, contact me.”
She handed him a business card. Professor Sarah Chen, Stanford University, Department of Electrical Engineering.
“I will,” Dojun said. “And professor—it’s an honor to meet you. Your work on adaptive throttling was foundational for my power consumption model.”
She smiled. “You read my 2003 paper?”
“I read everything.”
“I can tell.” She left, and Dojun stood holding two business cards—Choi Eunji’s from September and now Sarah Chen’s from June. Two doors, opened by the same work, leading to two different futures.
His phone was exploding again. Hana, Minjae, Seokho, Kim Taesik—all watching the conference live stream, all sending messages he couldn’t read through blurred eyes.
He stepped outside the ballroom into the San Diego afternoon. The sun was still blinding. The sky was still impossibly blue. Palm trees lined the convention center plaza, their fronds rustling in a warm breeze that carried the salt-tang of the Pacific Ocean.
He called his mother.
“How was it?” she asked.
“Good, Mom. Really good.”
“Did they clap?”
“They clapped.”
“Did anyone say you were too young?”
“No. They said the work was good.”
“Of course it’s good. My son’s work is always good.” A pause. “Did you eat the kimchi?”
“I ate the kimchi.”
“Good. Kimchi gives confidence. That’s science.”
“That’s definitely science.”
“Come home safely, Dojun-ah. The market is quiet without you. Mrs. Kang keeps asking when you’re coming back to fix her laptop.”
“Saturday, Mom. I’ll be there Saturday.”
“Good. I’ll make galbitang.”
She hung up. He stood in the San Diego sun and let the warmth soak into him—California warmth, so different from Seoul’s thin spring light, but warmth all the same.
On the flight home, somewhere over the Pacific Ocean, Dojun opened his laptop and stared at an empty file.
Not Bridge code. Not a research paper. Not a pitch deck or a business plan or an algorithm.
Just a blank file. A cursor blinking in the dark of the cabin.
He typed:
One year and three months since I woke up. The first volume is done.
Not a volume of a novel. A volume of a life. The introductions are made, the relationships are built, the first project is real. Bridge exists. The paper is presented. The people are here.
In my first life, the first volume ended differently. I was alone. Brilliant, successful, and alone. The company was growing but the person building it was shrinking. I didn’t visit my mother. I didn’t listen to my partner. I didn’t have friends—I had colleagues.
This time, the first volume ends like this:
A mother who uses a laptop to track her banchan sales and calls a mouse a “drunk squirrel.”
A partner who kissed me on a subway platform and designs for devices that don’t exist yet.
A rival who became a collaborator and is about to build his own company because he watched me build mine.
A friend who walked every path on campus and screamed when we won.
A mentor who doesn’t understand me but believes in me anyway.
A product that helps five hundred students organize their homework and two ajummas track their banchan.
A paper presented at the world’s best conference. A second paper accepted at another. A startup funded, growing, alive.
And a twenty-year-old man who is, slowly, learning that the most important code he will ever write is not the code that runs on machines. It’s the code that runs between people. The invisible connections, the kept promises, the Saturday visits, the 3 AM phone calls, the shared bowls of jjigae in a basement restaurant.
The first volume is done.
The second volume begins with a phone. A device that will change the world—again—and that Bridge is already designed for, though nobody knows it yet.
But that’s the next chapter. Tonight, somewhere over the Pacific, I’m just a kid flying home.
He saved the file. Closed the laptop. Leaned his seat back and looked out the window at the ocean of clouds below, lit orange by the setting sun, stretching in every direction like an endless, luminous floor.
The first time he had crossed this ocean, he had been thirty-two, flying to a conference, thinking about nothing but work. He hadn’t looked out the window. He hadn’t noticed the clouds. He hadn’t felt the particular wonder of being suspended between continents, between time zones, between the person you were and the person you were becoming.
Now he noticed. He noticed everything.
The plane carried him east, toward morning, toward Seoul, toward the people waiting for him. Hana with her sketchbook. Minjae with his dashboard. Seokho with his questions. Kim Taesik with his terrible coffee. His mother with her galbitang.
And in his pocket, a small container—empty now—that still smelled faintly of kimchi.
The clouds burned gold beneath him. The engines hummed their steady song. And Park Dojun, twenty-one years old, sixty-three years of memory, one year of second chances, closed his eyes and slept.
Not the exhausted sleep of a man running from his past. The peaceful sleep of a man running toward his future.
Volume one, complete.
The cursor blinks. The next line waits.