5 ms·

You’re missing the cost of engineering, for one thing. Also “host Uber at OVH” isn’t remotely realistic.

by rzz3 4y ago

HappyTypist 4y ago
Okay, OVH is not realistic, but at Uber scale, you can certainly roll your own data-centres and get costs lower than, or similar to OVH, even when including the cost of sysadmin and maintenance. Yes, it means you won't get all the shiny quality of life services offered by cloud providers, but you're in a _margin sensitive business_. Deal with it. Optimize every cost. Also, you don't need expensive engineers re-inventing the most basic things (I know Uber had a huge not-invented-here syndrome). Use the boring tools for the job. Only reinvent what is necessary. You don't need engineers practicing resume-driven-development.
- bthrn 4y ago
  Uber does use its own data centers.
onlyrealcuzzo 4y ago
For a simplified version of the app with arguably no worse user experience, it's not THAT farfetched. Uber has an absurd amount of logging & analytics. If the app was simply drop a pin, get a ride - it wouldn't be that crazy. Uber has 3.9M drivers world wide. There's probably very rarely more than 1M drivers active at any time. Probably less than 300k people looking for a ride at the vast majority of times. Assuming you can update the driver's location 1 time per minute - that's ~1.5B requests per day - less than 25k requests per second (including user bookings). That's like ~2TB of bandwidth per day. That's less than $200 per day. Almost everyone spends more than 5% of their cloud bill on bandwidth. Meaning, the rest of a drastically simplified (but nearly equally useful) Uber could run for ~$4000 per day in server expenses. That's a $1.4M / year data center. Uber has revenues of >$11B. They could be making a lot of money. They just aren't because they're spending AT LEAST 50x more on servers and product engineering than they NEED to. They paid for growth for a long time. They have a monopoly now. There's not a lot of growth left to get. At some point the axe will come down. Lyft is even worse. They're ~1/3rd the size.
- GauntletWizard 4y ago
  You're way overoptimistic in your expenses calculation - Maintaining a "real-time" app is 100x more complex than that. 100x, bringing their opex to hundreds of millions on billions of revenue. They should still be able to cut the fat and actually turn a handy profit, but they won't.
  - onlyrealcuzzo 4y ago
    No one needs real-time (above 1 minute) driver locations. Riders never see the driver's location before booking. Uber doesn't need the exact location to get a decent match. And anyway - they send the ride to several drivers and the lowest / first bidder wins.
    - lumost 4y ago
      In a major metro like Boston 1 minute further down the road could mean a 5 minute longer wait. I think your point stands however with 6x traffic increase to once every 10 seconds.
      - onlyrealcuzzo 4y ago
        You only need to update the location if the driver is moving. IIUC, most drivers are stagnant while they wait for rides. They're in the business of saving gas. Not cruising around idly while they wait for a ride. Sure - some drivers are finishing a trip nearby. But you know the route they are taking...
        noirbot 4y ago
        I'm not sure that's true. Almost every rideshare I've gotten in the last few years outside of an airport pickup has been in motion when they got assigned to me, and usually already on another ride. Even many of my airport rides were being picked up by someone who was dropping off someone else at the airport right before. You also don't know for-sure what route they're taking. Almost every driver I've had was using Waze or some other mapping app that usually took a different reasonable route than whatever Lyft was showing when I booked it.

You’re missing the cost of engineering, for one thing. Also “host Uber at OVH” isn’t remotely realistic.

by rzz3 4y ago

You’re missing the cost of engineering, for one thing. Also “host Uber at OVH” isn’t remotely realistic.

Okay, OVH is not realistic, but at Uber scale, you can certainly roll your own data-centres and get costs lower than, or similar to OVH, even when including the cost of sysadmin and maintenance. Yes, it means you won't get all the shiny quality of life services offered by cloud providers, but you're in a _margin sensitive business_. Deal with it. Optimize every cost. Also, you don't need expensive engineers re-inventing the most basic things (I know Uber had a huge not-invented-here syndrome). Use the boring tools for the job. Only reinvent what is necessary. You don't need engineers practicing resume-driven-development.

bthrn 4y ago
Uber does use its own data centers.

onlyrealcuzzo 4y ago

For a simplified version of the app with arguably no worse user experience, it's not THAT farfetched. Uber has an absurd amount of logging & analytics. If the app was simply drop a pin, get a ride - it wouldn't be that crazy. Uber has 3.9M drivers world wide. There's probably very rarely more than 1M drivers active at any time. Probably less than 300k people looking for a ride at the vast majority of times. Assuming you can update the driver's location 1 time per minute - that's ~1.5B requests per day - less than 25k requests per second (including user bookings). That's like ~2TB of bandwidth per day. That's less than $200 per day. Almost everyone spends more than 5% of their cloud bill on bandwidth. Meaning, the rest of a drastically simplified (but nearly equally useful) Uber could run for ~$4000 per day in server expenses. That's a $1.4M / year data center. Uber has revenues of >$11B. They could be making a lot of money. They just aren't because they're spending AT LEAST 50x more on servers and product engineering than they NEED to. They paid for growth for a long time. They have a monopoly now. There's not a lot of growth left to get. At some point the axe will come down. Lyft is even worse. They're ~1/3rd the size.

GauntletWizard 4y ago
You're way overoptimistic in your expenses calculation - Maintaining a "real-time" app is 100x more complex than that. 100x, bringing their opex to hundreds of millions on billions of revenue. They should still be able to cut the fat and actually turn a handy profit, but they won't.
- onlyrealcuzzo 4y ago
  No one needs real-time (above 1 minute) driver locations. Riders never see the driver's location before booking. Uber doesn't need the exact location to get a decent match. And anyway - they send the ride to several drivers and the lowest / first bidder wins.
  - lumost 4y ago
    In a major metro like Boston 1 minute further down the road could mean a 5 minute longer wait. I think your point stands however with 6x traffic increase to once every 10 seconds.
    - onlyrealcuzzo 4y ago
      You only need to update the location if the driver is moving. IIUC, most drivers are stagnant while they wait for rides. They're in the business of saving gas. Not cruising around idly while they wait for a ride. Sure - some drivers are finishing a trip nearby. But you know the route they are taking...
      - noirbot 4y ago
        I'm not sure that's true. Almost every rideshare I've gotten in the last few years outside of an airport pickup has been in motion when they got assigned to me, and usually already on another ride. Even many of my airport rides were being picked up by someone who was dropping off someone else at the airport right before. You also don't know for-sure what route they're taking. Almost every driver I've had was using Waze or some other mapping app that usually took a different reasonable route than whatever Lyft was showing when I booked it.