Chapter 29: Pisa
The Leaning Tower looked exactly like the postcards, which was simultaneously disappointing and comforting—proof that some things in the world were exactly as advertised.
Dojun and Seokho stood in the Piazza dei Miracoli on a blistering August afternoon, eating gelato and looking at the tower with the detached appreciation of two computer scientists who understood structural engineering well enough to know that the building should have collapsed centuries ago.
“Differential settlement of the foundation,” Seokho said, licking pistachio gelato with the same mechanical efficiency he applied to everything. “The soil on the south side is clay; the north side is sand. Uneven compression over eight hundred years. The corrective measures in the 1990s stabilized it, but the underlying physics hasn’t changed.”
“You researched the tower’s geotechnical history before coming to Italy?”
“I research everything before going anywhere. That’s not unusual.”
“Most people just take photos.”
“Most people don’t understand what they’re photographing.” He finished his gelato. “Our presentation is in three hours. Are you nervous?”
“Should I be?”
“SIGCOMM is smaller than ISCA but meaner. The networking community is territorial. They don’t like outsiders—especially outsiders from architecture who think they can solve networking problems.”
“We did solve a networking problem.”
“We solved it using architecture principles. The networking people will view that as an invasion. Expect hostile questions about whether our dual-mode model accounts for TCP congestion dynamics, because someone will try to find a failure case that invalidates the cross-domain application.”
“And does it? Account for TCP congestion?”
“Of course it does. I’m not an amateur.” He pulled a folded paper from his pocket—notes, handwritten, organized by anticipated question topic. “I’ve prepared responses for seventeen possible hostile questions. You should review them.”
“Seventeen?”
“Eighteen, actually. I added one this morning about flow-level fairness. A professor from UCL published a critique of dual-mode classification last month. She argues that separating predictable and unpredictable flows creates a priority inversion problem.”
“Does it?”
“Only if you implement the classifier naively, which we didn’t. Our interference-aware scheduling prevents priority inversion by dynamically rebalancing resources between modes.” He handed over the notes. “Read these. Memorize answers for questions seven through twelve—those are the ones I can’t answer as well as you can.”
Dojun took the notes. Seokho’s handwriting was as precise as his code—angular, consistent, with the kind of uniformity that suggested a mind that valued structure over expression.
“Seokho,” Dojun said. “Have you ever just… enjoyed a moment? Without preparing for it?”
“I’m enjoying this moment. I’m in Pisa, eating gelato, about to present at SIGCOMM. The fact that I’m also preparing doesn’t diminish the enjoyment. It enhances it. Preparation is how I experience joy.”
“That’s the saddest beautiful thing I’ve ever heard.”
“It’s efficient. Sadness and beauty should coexist for maximum emotional bandwidth.” The near-smile. “Let’s go. I want to rehearse the transition between your section and mine. The handoff on slide nine was clunky in our last practice.”
The SIGCOMM presentation went better than either of them expected—and Seokho expected a lot.
Their paper was scheduled in the “Cross-Layer Design” track, a session that attracted researchers interested in breaking down the traditional boundaries between network layers. The audience was seventy people—smaller than ISCA, but more specialized, with the focused attention of a community that cared deeply about its domain.
Seokho presented the networking foundations—the bimodal distribution of network traffic, the limitations of single-mode classification, the experimental methodology. His delivery was dry, precise, and devastatingly thorough. He answered the first two questions—about dataset selection and simulation parameters—with the kind of detail that made the questioners nod and sit back.
Then Dojun took over for the theoretical framework—the dual-mode prediction model, adapted from branch prediction to traffic classification. He explained the cross-domain analogy with the careful clarity of someone bridging two worlds that rarely talked to each other.
“Processor branches and network flows share a fundamental property,” he said. “Both are generated by systems with mixed behavior—some components are regular and predictable, others are irregular and stochastic. The dual-mode approach works because it respects this dichotomy instead of trying to force a single model onto both types.”
The UCL professor—the one Seokho had warned about—raised her hand during Q&A.
“Your dual-mode classifier assumes that flow behavior is stable over time. But TCP flows change behavior during congestion events—a previously predictable flow becomes unpredictable. How does your model handle dynamic reclassification?”
“The classifier runs continuously, not once,” Dojun said. “Every classification window—thirty seconds in our implementation—the model re-evaluates each flow. A flow that transitions from regular to irregular is reclassified at the next window boundary. The cost of reclassification is bounded by the window size.”
“Thirty seconds is a long window for congestion response.”
“For real-time applications, yes. We acknowledge this limitation in the paper—section 5.3. For our target use case, which is software-defined network management, thirty-second windows are acceptable because the SDN controller operates on longer timescales. Real-time congestion response remains a separate mechanism.”
The professor made a note. Not hostile—thoughtful. The kind of note that meant she was considering a follow-up paper rather than a rebuttal.
After the session, she approached them.
“Your cross-domain analogy is interesting,” she said. “I’m skeptical about the window size, but the theoretical framework is sound. I’d like to cite your work in my next paper on flow classification taxonomy.”
“We’d be honored,” Seokho said.
“Don’t be honored. Be cited. It’s more useful.” She handed them each a business card and left.
Seokho examined his card. “Professor Margaret Wu, University College London. She’s one of the top five researchers in network flow classification worldwide.”
“And she wants to cite us.”
“She wants to engage with our work, which may include citing us or dismantling our model. Either way, it means the paper matters.” He put the card in his wallet with the precise care of someone filing a valuable asset. “Good presentation, Park. Your ‘I don’t know’ on the reclassification latency question was convincing.”
“I didn’t say ‘I don’t know.'”
“You said ‘we acknowledge this limitation,’ which is academic for ‘I don’t know the answer yet but I’ve thought about it.’ Same thing, better optics.” He checked his watch. “We have four hours before the conference dinner. I want to show you something.”
“What?”
“Pisa’s other tower. The one that doesn’t lean. Nobody visits it, but the architecture is actually more interesting.”
They walked through Pisa’s narrow streets as the afternoon cooled into evening. The city was smaller than Dojun expected—intimate, almost village-like, with terracotta rooftops and winding alleys that opened onto unexpected piazzas. The Arno River moved slowly through the center, its water green-brown and reflective.
“Nova Systems,” Dojun said as they walked. “How’s it going?”
“Slowly. I filed incorporation last month. I have a business plan, a technical architecture, and zero funding.” Seokho walked with the long, efficient stride of someone who treated walking as transportation, not recreation. “The Korean investment market doesn’t understand cloud infrastructure. I’ve pitched to four firms. All four said the same thing: ‘Why would companies pay for remote servers when they can buy their own?'”
“Because buying your own servers is expensive, slow, and requires expertise that most companies don’t have.”
“I know that. You know that. Amazon knows that—AWS is growing at 100% year over year. But Korean investors can’t see past the hardware mindset. They think computing is something you own, not something you rent.”
“That will change.”
“When?”
“Sooner than they think. The mobile revolution will create demand for elastic infrastructure. When millions of smartphone users start downloading apps that need server backends, the companies building those apps won’t want to buy and manage their own servers. They’ll want to rent compute power on demand. That’s your market.”
Seokho stopped walking. “You’re doing it again.”
“Doing what?”
“Predicting the future with a specificity that transcends analysis. ‘Millions of smartphone users.’ ‘Elastic infrastructure.’ ‘Compute power on demand.’ These aren’t guesses, Park. These are descriptions of a world that doesn’t exist yet, delivered with the confidence of someone who’s already lived in it.”
The Italian evening was warm and still. A church bell rang somewhere. A scooter buzzed past on a nearby street.
“Maybe I’m just very good at extrapolation,” Dojun said.
“Maybe. Or maybe you’re the most unusual person I’ve ever met, and after two and a half years of trying to figure out why, I’ve decided to stop trying and just benefit from it.” He resumed walking. “Whatever you see that I don’t—keep seeing it. And keep telling me. I’ll build the infrastructure. You build the products. Together, we’ll be right about things that nobody else believes yet.”
“That sounds like a partnership.”
“It sounds like a competitive advantage.” The near-smile. “Same thing, better framing.”
They found the other tower—a bell tower attached to the Church of San Nicola, perfectly vertical, unvisited, with stone walls that had stood for seven centuries without leaning an inch. Seokho studied its construction with the same analytical intensity he brought to algorithms.
“This one did it right,” he said. “Proper foundation. Uniform soil analysis. No differential settlement. It’s not famous because it’s not broken. Nobody visits the tower that works correctly.”
“There’s a lesson in that.”
“The lesson is that failure is more interesting than success. The Leaning Tower attracts millions because it almost fell down. This tower attracts nobody because it did exactly what it was designed to do.” He paused. “Bridge is the tower that works. Nobody will write articles about a task organizer that functions smoothly. They’ll write articles about the companies that fail spectacularly. Your job is to make sure you’re boring—successfully, quietly, invisibly boring.”
“Invisible technology.”
“Your phrase. But it applies to infrastructure too. The best cloud platform is the one nobody thinks about. The servers just work. The data just flows. The compute just scales.” He looked up at the tower. “I want to build the tower that nobody visits because it never leans.”
“And I want to build the software that nobody notices because it just works.”
“Then we’re both chasing invisibility.” Seokho extended his hand. “To being boring together.”
Dojun shook it. “To being boring together.”
They stood in the shadow of a perfect, forgotten tower in a small Italian city, two Korean twenty-somethings who had just presented a paper to seventy of the world’s best networking researchers, and who were both, in their own ways, trying to build things that would change the world by disappearing into it.
On the flight home, Dojun’s phone loaded a backlog of messages from the sixteen hours he’d been in transit and out of range.
Hana: How was Pisa? Did the tower lean? Did Seokho lecture you about geotechnical engineering? More importantly — the App Store opened yesterday. I submitted Bridge Mobile at 9 AM. We’re in the review queue. Estimated approval: 3-5 days.
His heart rate spiked. The App Store. July 2008. Day one.
You submitted without me?
You were on a plane. The App Store opened. I wasn’t going to wait. The build was ready. Taeyoung tested it six times. Minjae verified the data pipeline. I reviewed every pixel. It was ready.
And if it wasn’t?
Then we’d fix it and resubmit. But it was ready. I know when our product is ready, Dojun. That’s my job.
She was right. Of course she was right. The product was ready because Hana had made it ready—because she had spent months polishing every interaction, reviewing every animation, testing every edge case with the uncompromising attention to detail that made her simultaneously the best designer he’d ever worked with and the most challenging partner he’d ever had.
You’re right. I’m sorry.
Don’t be sorry. Be excited. We’re about to be in the App Store. THE APP STORE. Day one. First wave. I can’t breathe.
Breathe. Then tell me everything.
Come to the office first. I want to show you the final build on an actual iPhone. I borrowed one from Professor Oh’s mobile lab. IT’S BEAUTIFUL.
Another message, from Minjae: Hana submitted the app while you were gone. I want you to know that I tried to convince her to wait for you and she said, and I quote, “The App Store doesn’t wait for people to land.” She’s terrifying and I love working for her.
And from his mother: Mrs. Kang says her grandson saw you on TV. You were at a computer meeting in Italy? Italy! That’s where they make the pasta. Did you bring pasta? Call me when you land.
He called her. She answered on the first ring.
“Did you bring pasta?”
“I brought olive oil. And some cheese.”
“Cheese from Italy! Fancy. Is it good?”
“It’s Parmigiano-Reggiano. Twenty-four months aged.”
“I don’t know what that means but it sounds expensive. Save it for Saturday. I’ll make japchae with Italian cheese. Fusion cuisine.”
“Mom, you can’t put Parmesan on japchae.”
“I can put anything on japchae. I’ve been making japchae for thirty years. Don’t tell me what goes on japchae.” She paused. “How was Italy?”
“Beautiful. We presented the paper. It went well.”
“Of course it went well. My son’s papers always go well.” Another pause. “Dojun-ah. You’ve been to America and Italy in the same year. When I was your age, I’d never left Seoul.”
“You built a business that’s lasted thirty years without leaving Seoul. That’s more impressive than any trip.”
“Don’t compare. My market stall and your computer conferences are different worlds.”
“They’re not as different as you think.”
“Silly boy.” But her voice was warm. “Come Saturday. Bring the cheese. And bring Hana. She’s been coming every other week now—she helps me with the laptop and argues with Mrs. Kang about sock display strategies. I like her, Dojun-ah. Very much.”
“I like her too, Mom.”
“I know. I’ve known since the first time you said her name. A mother hears these things.” She hung up with her customary abruptness—conversations ended when they ended, no need for prolonged goodbyes.
Dojun pocketed his phone and stepped out of Incheon Airport into the August night. Seoul was humid and hot, the air thick enough to wear. Traffic roared on the expressway. The city lights reflected off low clouds, casting an orange glow over everything.
Home. Korea. The place where a banchan vendor’s son was building a software company that was about to launch on the most important platform in the history of mobile computing.
Somewhere in Cupertino, an Apple review team was looking at Bridge Mobile’s submission. Somewhere in the Bridge office, Hana was probably still working, refining something that nobody but her would notice. Somewhere in Daejeon, Seokho was probably already writing the first line of code for Nova Systems.
The future was arriving, one day at a time, in the particular Korean way—not with Silicon Valley’s bombastic disruption, but with the quiet, persistent determination of people who showed up every morning, did the work, and went home to eat their mother’s cooking.
Bridge Mobile. Day one of the App Store. The beginning of volume two.
He hailed a taxi and headed toward Gwanak-gu. There was an iPhone in the office with his product on it, and he wanted to see it.