6 ms·

With a round that big, it looks like Dotcloud is going to buy their own hardware and become a full bore hosting company, which is fantastic, if it can convince

With a round that big, it looks like Dotcloud is going to buy their own hardware and become a full bore hosting company, which is fantastic, if it can convince people to move out of the EC2 slum. Either that, or the founders took too much money and diluted themselves beyond reason.

fuzzmeister 16y ago
Just curious, why do you call EC2 a "slum"? We've had nothing but good experiences with it.
- tedjdziuba 16y ago
  All of AWS, really. My experiences with EC2 were abysmal: unpredictable CPU and disk performance, uselessly slow disk I/O, instances dropping off the face of the earth at least once a week. Reddit's recent downtime is a good indicator of how sloppy AWS as a whole really is.
  - gsharma 16y ago
    I agree with slow disk I/O, but how do you think Netflix is doing so well with EC2? Adding more machines than they need to?
    - tt 16y ago
      They put in great engineering effort to ensure high availability despite random instance failures. I really like their Chaos Monkey. See: http://techblog.netflix.com/2010/12/5-lessons-weve-learned-using-aws.html http://techblog.netflix.com/2010/12/5-lessons-weve-learned-u...
      - gfodor 16y ago
        One could argue this is how all systems should be designed anyway.
      - dongle 16y ago
        It's cool that Netflix pulled that off, but it seems like a company (like Dotcloud) that could prevent you from building infrastructure on top of infrastructure would be a boon.
    - tedjdziuba 16y ago
      I don't have any data points about Netflix's migration other than their public statements about it, which undoubtedly would conceal any major issues they are struggling with. From what they have said publicly: - They had to re-architect their entire system To me, this seems like a poor business decision. I don't know the specifics of their CapEx, but it seems unlikely that the money saved by cotenancy at Amazon is worth the risk of changing known-working, business critical code. - Machine failure is such a strong expectation that they have a system that purposefully brings machines down. While I applaud the pedantic reasoning behind this, I suspect it's driven more by technical masturbation than business needs.
    - cpeterso 16y ago
      As a high-profile EC2 client, I imagine Amazon gives Netflix tons of "complementary" support. A Netflix failure on EC2 would be a big black eye to Amazon PR.
    - lsc 16y ago
      the outages@outages.org mailing list says that Netflix was down today. I don't know if it's ec2 related or not... I was actually hoping someone might have details. https://puck.nether.net/pipermail/outages/2011-March/002786.html https://puck.nether.net/pipermail/outages/2011-March/002786....
shykes 16y ago
Co-founder here. We're considering our infrastructure options, but our #1 priority is to make developer's lives easier, not to buy servers. As for dilution - let's just say we feel very good about our current ownership.
- tedjdziuba 16y ago
  Thanks for your reply. If you're not going to buy your own metal, then with that much cash, you're probably going to hire an army of systems administrators to be oncall for your customers, eh?
  - spahl 16y ago
    More like an army of engineers to automate the hell out of it:-)
- phlux 16y ago
  I'm always intrigued by services that rely on an infrastructure layer you don't control. For example - you are betting that, say, AWS will put in both the best of breed hardware for your particular service - and that also they will manage it, set it up correctly / to your needs, and adjust to your requests. Yet as a small player - you likely have little negotiating power with them, so how much visibility do you have to their implementations? Further, there are HW caching and high speed memory layer technologies (FusionIO, Violin memory) that services like yours can greatly benefit from - yet you may not have any option to use such technologies due to reliance on a hosting vendor relationship. What are your thoughts here? I think that a hybrid model could be deployed where you define an architecture that is the best for your company and that takes advantages of the strengths of hosting -- for example - you don't want to maintain all the switching, routing and potentially storage gear - but maybe you could define the cache layer and install an maintain that with all your traffic routing through it which front ends to the more traditional offerings of EC2/AWS?

fuzzmeister 16y ago

Just curious, why do you call EC2 a "slum"? We've had nothing but good experiences with it.

tedjdziuba 16y ago
All of AWS, really. My experiences with EC2 were abysmal: unpredictable CPU and disk performance, uselessly slow disk I/O, instances dropping off the face of the earth at least once a week. Reddit's recent downtime is a good indicator of how sloppy AWS as a whole really is.
- gsharma 16y ago
  I agree with slow disk I/O, but how do you think Netflix is doing so well with EC2? Adding more machines than they need to?
  - tt 16y ago
    They put in great engineering effort to ensure high availability despite random instance failures. I really like their Chaos Monkey. See: http://techblog.netflix.com/2010/12/5-lessons-weve-learned-using-aws.html http://techblog.netflix.com/2010/12/5-lessons-weve-learned-u...
    - gfodor 16y ago
      One could argue this is how all systems should be designed anyway.
    - dongle 16y ago
      It's cool that Netflix pulled that off, but it seems like a company (like Dotcloud) that could prevent you from building infrastructure on top of infrastructure would be a boon.
  - tedjdziuba 16y ago
    I don't have any data points about Netflix's migration other than their public statements about it, which undoubtedly would conceal any major issues they are struggling with. From what they have said publicly: - They had to re-architect their entire system To me, this seems like a poor business decision. I don't know the specifics of their CapEx, but it seems unlikely that the money saved by cotenancy at Amazon is worth the risk of changing known-working, business critical code. - Machine failure is such a strong expectation that they have a system that purposefully brings machines down. While I applaud the pedantic reasoning behind this, I suspect it's driven more by technical masturbation than business needs.
  - cpeterso 16y ago
    As a high-profile EC2 client, I imagine Amazon gives Netflix tons of "complementary" support. A Netflix failure on EC2 would be a big black eye to Amazon PR.
  - lsc 16y ago
    the outages@outages.org mailing list says that Netflix was down today. I don't know if it's ec2 related or not... I was actually hoping someone might have details. https://puck.nether.net/pipermail/outages/2011-March/002786.html https://puck.nether.net/pipermail/outages/2011-March/002786....

shykes 16y ago

Co-founder here. We're considering our infrastructure options, but our #1 priority is to make developer's lives easier, not to buy servers. As for dilution - let's just say we feel very good about our current ownership.

tedjdziuba 16y ago
Thanks for your reply. If you're not going to buy your own metal, then with that much cash, you're probably going to hire an army of systems administrators to be oncall for your customers, eh?
- spahl 16y ago
  More like an army of engineers to automate the hell out of it:-)
phlux 16y ago
I'm always intrigued by services that rely on an infrastructure layer you don't control. For example - you are betting that, say, AWS will put in both the best of breed hardware for your particular service - and that also they will manage it, set it up correctly / to your needs, and adjust to your requests. Yet as a small player - you likely have little negotiating power with them, so how much visibility do you have to their implementations? Further, there are HW caching and high speed memory layer technologies (FusionIO, Violin memory) that services like yours can greatly benefit from - yet you may not have any option to use such technologies due to reliance on a hosting vendor relationship. What are your thoughts here? I think that a hybrid model could be deployed where you define an architecture that is the best for your company and that takes advantages of the strengths of hosting -- for example - you don't want to maintain all the switching, routing and potentially storage gear - but maybe you could define the cache layer and install an maintain that with all your traffic routing through it which front ends to the more traditional offerings of EC2/AWS?