9 ms·

Many people use AWS because everyone uses AWS. Many of my clients have no need for AWS but still use it, at least until the VC money runs out. Usually then I ha

Many people use AWS because everyone uses AWS. Many of my clients have no need for AWS but still use it, at least until the VC money runs out. Usually then I have to go first, then servers are moving to somewhere cheaper when a new CFO comes in.

lvh 6y ago
It's possible that "many" firms do this, but given AWS' growth numbers that would imply they don't really spend meaningful amounts to begin with. Counterpoint: I've seen a great many false economies with people trying to go on-prem and do alternative hosting because they don't think the AWS premium for e.g. GPU instances is worth it. I don't think that has generally worked out well.
- KingOfCoders 6y ago
  "It's possible that "many" firms do this, but given AWS' growth numbers that would imply they don't really spend meaningful amounts to begin with." I could spend a million EUR a year on AWS without the need for most of the services of AWS? But it YMMV and 1M EUR/y is not meaningful, perhaps we differ there.
- qaq 6y ago
  There is no good data to prove either statement. You could def prove that for some large scale onprem for a given type of problem is cheeper than the cloud.
- freeone3000 6y ago
  The AWS premium for GPU instances is absolutely not worth it. You don't hear about people running local GPU compute clusters because it's not newsworthy -- it's obvious. Put a few workstations behind a switch, fire up torch.distributed, you're done. And after two months, you've beaten the AWS spend for the same amount of GPU compute, even if only 50% of the time is spent training. Timesharing is done with shell accounts and asking nicely. You do not need the huge technical complexity of the cloud: it gets in the way, as well as costing more!
  - srtjstjsj 6y ago
    What if you want 10x the GPU for one month to build a model?
    - freeone3000 6y ago
      That's the only scenario I can think where it comes out clearly in favor of AWS - you've tested your model in the small on Colab, you're confident you'll need only a few training runs, you can schedule them in us-east, and you can inference on CPU, and you won't need to rebuild for another eight months (when purchased cards become outdated). It's not an impossible scenario... But imagine the sort of company that trains their own model instead of using a huggingface refinement or an off-the-shelf redistillation. (These can be done reasonably on an average gaming PC, no need for a cluster.) Such a company has expensive human resources. They bothered to get a data scientist and at least a research engineer, if not a full researcher. Were they hired on six-month contract as well? This is a huge expense, so it must be an important differentiator to have built a custom model -- and it's a one-and-done? I don't see it. I think it's going to be an ongoing project, or it shouldn't have been approved in the first place.
Cthulhu_ 6y ago
From AWS' own marketing propaganda: How much would said clients have to invest in people and hardware otherwise? What if their application becomes an overnight success and needs to scale up fast? Sure, if it's an established company then using their own hardware (and people to set up and manage it) might make sense; iirc Dropbox is a fairly recent big player that made that move. But otherwise it's a big upfront investment to make, and you can't know if it'll pay itself back. So sure, AWS can be 2x or more as expensive as renting servers at a company like OVH or building your own datacenter, but it's paid by the minute if need be, not all in one go. If your startup or its VC money runs out in six months at least you can quickly scale down and pull out.
- bsenftner 6y ago
  For my own startup, I built a small cluster of 17 servers for just beneath $55K, and that had a month-to-month expense of $600 placed in a co-lo. In comparison, the same setup at AWS would be $96K per month. And it is not hard, easy in many ways. Do not be fooled, the cloud companies are peddling is an expensive scam.
  - Retric 6y ago
    Cloud companies are useful as long as you want what their selling. The best case is needing say 2 TB of ram for some workload or test that only going to take a few hours. Or something like the Olympics where 95% of your demand comes in a predictable spike across a few days.
    - chasd00 6y ago
      very true, the original selling point for Cloud was instant upgrade/downgrade as needed. That was the original amazing thing, dials that said RAM and CPU you could turn up or turn down.
  - ti_ranger 6y ago
    > For my own startup, I built a small cluster of 17 servers for just beneath $55K, and that had a month-to-month expense of $600 placed in a co-lo. In comparison, the same setup at AWS would be $96K per month. Why would you build exactly the same setup in AWS as for on-prem, unless your objective is to (dishonestly) show that on-prem is cheaper? Lift-and-shift-to-the-cloud is known to be more expensive, because you aren't taking advantage of the features available to you which would allow you to reduce your costs.
WrtCdEvrydy 6y ago
AWS also gives you a lot of cost savings for using Spot and signing contracts with minimum spends. Only a small shop pays full price for anything.
- shiftpgdn 6y ago
  Okay but then you have to engineer your application around interuptable spot instances. You're also making a 36 month commitment when you sign that contract (generally buying the hardware for Amazon.)
  - WrtCdEvrydy 6y ago
    > you have to engineer your application around interuptable spot instances This is where your ALBs and ASGs come in. If your app doesn't use local writing and you can shift your caching to a shared cache, the cost savings are good.
ehnto 6y ago
I work in an ecommerce agency environment and it can be quite frustrating. Every client, no matter how small, has been sold the spiel and had some guy come through and build a boutique cloud infrastructure, CI pipeline and development VM. As if traffic will increase 1000% by next week with no prior warning. It just doesn't happen like that. I spend half my time debugging Hyperscale Bullshit™ that the client never needed and that failed the moment the devops guy left. If you want this stuff, you need a long term tech team, and a devops member on hand. Every time you hand your project around this stuff costs you hours. You also need to tell your devs to use the devops guy like tech support. They built this rickety tower of obtuse config files, make the devops guy fix it and don't let your devs spin their wheels for 4 hours before their assets compile on the Super Cool Distributed Docker VM microservice stack of teetering jenga blocks.
- michaelbuckbee 6y ago
  I did work for a YC startup that replaced my ~$100/mo Heroku app with a $5,000/mo AWS stack that nobody in the place knew how to manage. Their site was so ridiculously low traffic that I did some napkin math and figured out they were paying around $0.10 per web request.
  - jrochkind1 6y ago
    Huh, making a manual AWS stack 50 times more expensive than a heroku setup seems like a real trick to me! I'm used to heroku being more expensive than running it yourself -- as I would expect, since you're paying them for a lot of management.
    - user5994461 6y ago
      Heroku is both incredibly cheap and incredibly expensive. It's only $7/month to deploy a web application. More if you want some of the paid features and a database instance. It all works out of the box with instant deployment out of the box, it's fantastic. Then it suddenly goes $50 more per gigabyte of RAM, which makes it massively expensive for any serious workload. It's crazy hot much they can try to charge, makes AWS looks like a bunch of clowns in comparison.
Spivak 6y ago
At my company we colo at a few local datacenters and have do deal with a huge amount of pressure from our investors as to why we're not using AWS. The points that always seem to come up: * AWS is a known quantity and it's easier to evaluate our business with it. * AWS provides "outage damage control" because AWS outages make the news and customers are more understanding. When our ISP has issues it just looks bad on us. * Our company doesn't look as innovative because we're not cloud. Bleh. Our app is compute, storage, and data transfer heavy but switching to AWS being a, literally, 10x cost for us is apparently not enough a good enough answer.
- mwcampbell 6y ago
  How do you do failover if a server fails or if connectivity to one of those datacenters is lost? With AWS I could just set up a multi-availability-zone RDS deployment for the database and an auto-scaling group for the web tier and be confident that AWS will recover the system from most failures. To me, that is the major selling point of any of the hyperscale cloud providers.
  - KingOfCoders 6y ago
    "connectivity to one of those datacenters is lost" Anecdotal, but AWS had more global problems than the triple connected data centers I've used in twenty years. I suffered through many rough times, data center connection or power problems were not (very seldom) one of those. Most of the problems came from apps that we didn't build well to scale or that had bugs (most frequent cause of problems and sitedowns).
  - HelloNurse 6y ago
    > With AWS I could just set up a multi-availability-zone RDS deployment for the database and an auto-scaling group for the web tier and be confident that AWS will recover the system from most failures "Confident"? "Most" failures? Are you merely hopeful that the probability of a bad failure is low, or are you able to test the AWS resiliency techniques you mention and to ensure that they stay working? At what cost?

lvh 6y ago

It's possible that "many" firms do this, but given AWS' growth numbers that would imply they don't really spend meaningful amounts to begin with. Counterpoint: I've seen a great many false economies with people trying to go on-prem and do alternative hosting because they don't think the AWS premium for e.g. GPU instances is worth it. I don't think that has generally worked out well.

KingOfCoders 6y ago
"It's possible that "many" firms do this, but given AWS' growth numbers that would imply they don't really spend meaningful amounts to begin with." I could spend a million EUR a year on AWS without the need for most of the services of AWS? But it YMMV and 1M EUR/y is not meaningful, perhaps we differ there.
qaq 6y ago
There is no good data to prove either statement. You could def prove that for some large scale onprem for a given type of problem is cheeper than the cloud.
freeone3000 6y ago
The AWS premium for GPU instances is absolutely not worth it. You don't hear about people running local GPU compute clusters because it's not newsworthy -- it's obvious. Put a few workstations behind a switch, fire up torch.distributed, you're done. And after two months, you've beaten the AWS spend for the same amount of GPU compute, even if only 50% of the time is spent training. Timesharing is done with shell accounts and asking nicely. You do not need the huge technical complexity of the cloud: it gets in the way, as well as costing more!
- srtjstjsj 6y ago
  What if you want 10x the GPU for one month to build a model?
  - freeone3000 6y ago
    That's the only scenario I can think where it comes out clearly in favor of AWS - you've tested your model in the small on Colab, you're confident you'll need only a few training runs, you can schedule them in us-east, and you can inference on CPU, and you won't need to rebuild for another eight months (when purchased cards become outdated). It's not an impossible scenario... But imagine the sort of company that trains their own model instead of using a huggingface refinement or an off-the-shelf redistillation. (These can be done reasonably on an average gaming PC, no need for a cluster.) Such a company has expensive human resources. They bothered to get a data scientist and at least a research engineer, if not a full researcher. Were they hired on six-month contract as well? This is a huge expense, so it must be an important differentiator to have built a custom model -- and it's a one-and-done? I don't see it. I think it's going to be an ongoing project, or it shouldn't have been approved in the first place.

Cthulhu_ 6y ago

From AWS' own marketing propaganda: How much would said clients have to invest in people and hardware otherwise? What if their application becomes an overnight success and needs to scale up fast? Sure, if it's an established company then using their own hardware (and people to set up and manage it) might make sense; iirc Dropbox is a fairly recent big player that made that move. But otherwise it's a big upfront investment to make, and you can't know if it'll pay itself back. So sure, AWS can be 2x or more as expensive as renting servers at a company like OVH or building your own datacenter, but it's paid by the minute if need be, not all in one go. If your startup or its VC money runs out in six months at least you can quickly scale down and pull out.

bsenftner 6y ago
For my own startup, I built a small cluster of 17 servers for just beneath $55K, and that had a month-to-month expense of $600 placed in a co-lo. In comparison, the same setup at AWS would be $96K per month. And it is not hard, easy in many ways. Do not be fooled, the cloud companies are peddling is an expensive scam.
- Retric 6y ago
  Cloud companies are useful as long as you want what their selling. The best case is needing say 2 TB of ram for some workload or test that only going to take a few hours. Or something like the Olympics where 95% of your demand comes in a predictable spike across a few days.
  - chasd00 6y ago
    very true, the original selling point for Cloud was instant upgrade/downgrade as needed. That was the original amazing thing, dials that said RAM and CPU you could turn up or turn down.
- ti_ranger 6y ago
  > For my own startup, I built a small cluster of 17 servers for just beneath $55K, and that had a month-to-month expense of $600 placed in a co-lo. In comparison, the same setup at AWS would be $96K per month. Why would you build exactly the same setup in AWS as for on-prem, unless your objective is to (dishonestly) show that on-prem is cheaper? Lift-and-shift-to-the-cloud is known to be more expensive, because you aren't taking advantage of the features available to you which would allow you to reduce your costs.

WrtCdEvrydy 6y ago

AWS also gives you a lot of cost savings for using Spot and signing contracts with minimum spends. Only a small shop pays full price for anything.

shiftpgdn 6y ago
Okay but then you have to engineer your application around interuptable spot instances. You're also making a 36 month commitment when you sign that contract (generally buying the hardware for Amazon.)
- WrtCdEvrydy 6y ago
  > you have to engineer your application around interuptable spot instances This is where your ALBs and ASGs come in. If your app doesn't use local writing and you can shift your caching to a shared cache, the cost savings are good.

ehnto 6y ago

I work in an ecommerce agency environment and it can be quite frustrating. Every client, no matter how small, has been sold the spiel and had some guy come through and build a boutique cloud infrastructure, CI pipeline and development VM. As if traffic will increase 1000% by next week with no prior warning. It just doesn't happen like that. I spend half my time debugging Hyperscale Bullshit™ that the client never needed and that failed the moment the devops guy left. If you want this stuff, you need a long term tech team, and a devops member on hand. Every time you hand your project around this stuff costs you hours. You also need to tell your devs to use the devops guy like tech support. They built this rickety tower of obtuse config files, make the devops guy fix it and don't let your devs spin their wheels for 4 hours before their assets compile on the Super Cool Distributed Docker VM microservice stack of teetering jenga blocks.

michaelbuckbee 6y ago
I did work for a YC startup that replaced my ~$100/mo Heroku app with a $5,000/mo AWS stack that nobody in the place knew how to manage. Their site was so ridiculously low traffic that I did some napkin math and figured out they were paying around $0.10 per web request.
- jrochkind1 6y ago
  Huh, making a manual AWS stack 50 times more expensive than a heroku setup seems like a real trick to me! I'm used to heroku being more expensive than running it yourself -- as I would expect, since you're paying them for a lot of management.
  - user5994461 6y ago
    Heroku is both incredibly cheap and incredibly expensive. It's only $7/month to deploy a web application. More if you want some of the paid features and a database instance. It all works out of the box with instant deployment out of the box, it's fantastic. Then it suddenly goes $50 more per gigabyte of RAM, which makes it massively expensive for any serious workload. It's crazy hot much they can try to charge, makes AWS looks like a bunch of clowns in comparison.

Spivak 6y ago

At my company we colo at a few local datacenters and have do deal with a huge amount of pressure from our investors as to why we're not using AWS. The points that always seem to come up: * AWS is a known quantity and it's easier to evaluate our business with it. * AWS provides "outage damage control" because AWS outages make the news and customers are more understanding. When our ISP has issues it just looks bad on us. * Our company doesn't look as innovative because we're not cloud. Bleh. Our app is compute, storage, and data transfer heavy but switching to AWS being a, literally, 10x cost for us is apparently not enough a good enough answer.

mwcampbell 6y ago
How do you do failover if a server fails or if connectivity to one of those datacenters is lost? With AWS I could just set up a multi-availability-zone RDS deployment for the database and an auto-scaling group for the web tier and be confident that AWS will recover the system from most failures. To me, that is the major selling point of any of the hyperscale cloud providers.
- KingOfCoders 6y ago
  "connectivity to one of those datacenters is lost" Anecdotal, but AWS had more global problems than the triple connected data centers I've used in twenty years. I suffered through many rough times, data center connection or power problems were not (very seldom) one of those. Most of the problems came from apps that we didn't build well to scale or that had bugs (most frequent cause of problems and sitedowns).
- HelloNurse 6y ago
  > With AWS I could just set up a multi-availability-zone RDS deployment for the database and an auto-scaling group for the web tier and be confident that AWS will recover the system from most failures "Confident"? "Most" failures? Are you merely hopeful that the probability of a bad failure is low, or are you able to test the AWS resiliency techniques you mention and to ensure that they stay working? At what cost?