6 ms·

Interesting idea, and I wish them the best of luck. However this line: > _Further, as cloud-borne SaaS companies mature from being strictly growth focused to b

Interesting idea, and I wish them the best of luck. However this line: > _Further, as cloud-borne SaaS companies mature from being strictly growth focused to being more margin focused, it seems likely that more will consider buying machines instead of always renting them_ Is not something I think to be true from my experience of being in a large company that has moved away from on-prem hosting to cloud providers. Cloud is cheaper. A lot cheaper, especially if using managed services (with the proper arch! With the exception of logging which is surprisingly expensive). The hardware costs of rolling your own hosting is not the expensive part.

jonny_eh 7y ago
Netflix famously moved their infrastructure from their own data centres to AWS.
- twoodfin 7y ago
  Though isn’t that the “commodity” part of their infrastructure? They’re presumably still spending like gangbusters on their “on premises” ISP-local hardware & software. I imagine that’s the extreme end of the market Oxide is targeting: Provide a HW/SW platform to build your differentiation on, because the capabilities and value (& moat) aren’t there in the commodity cloud.
  - hinkley 7y ago
    I really think this shouldn't be an either-or thing, most especially for established companies, but to some extent for growing ones as well. If you run a data center in the same city/cities as your engineering strengths are located, that doesn't necessarily give you the geographical diversity you need for many companies. But if you don't run any data centers, you're blind to some of the cost structures, and architectural limitations of the system, and you can become soft. I was going to make a simile to Google and Facebook having their own hardware divisions, but I don't really need to, because they are planning to make cloud-friendly custom hardware for data centers. This seems like an area that the incumbents should be all over but I can't recall the last time I read of innovations from them. Which means there is space for someone new to establish a toe-hold. If I were based in Chicago I'd want a data center in Chicago, and Cloud Servers on the West Coast, (and Europe, etc as applicable). But the lock-in situation is untenable to me right now. It's pretty easy to end up deploying multiple solutions for the same situation. That just complicates reasoning about the system. It bears a resemblance to the Lavaflow antipattern and I can tell you that either can be no fun at all.
    - hinkley 7y ago
      I'm not sure why I chose the phrase 'planning to' about custom hardware, when we know that they've been doing this for a very long time, and FB in particular has not only published designs but multiple cycles of amendments to those designs.
- outside1234 7y ago
  They also have the size and early mover status where they probably inked a pretty sweet deal to do that.
StacyRoberts 7y ago
The expensive part is the capacity per dollar. It's not cloud vs on prem. The real value play is colocation. Imagine operational costs at $100/mo for a terabyte of RAM and 32 cores on unmetered bandwidth compared to the equivalent at aws. Most sv startups don't care about operational costs but they do care about capital costs of buying the hardware. Primarily due to the very short term focus. I think their own resumes are important too vs their employers' long term viability so everyone pushes aws or azure etc because it's hot tech. Thinking long term is not a silicon valley thing. It's a totally different mindset.
- kbenson 7y ago
  At the very lowest end, colocation is a clear winner in value, even up to a few servers, it stands up well. Once you have multiple servers, and need to start load-balancing some sets of servers to deal with load spikes, cloud-based infrastructure starts to have some benefits that make it competitive and possible cheaper in some instances, because of load scaling and ease of turn-up of new systems. If you're increasing load by 20% or more a month, you likely will have major problems provisioning servers fast enough unless you gamble that trend will increase for a year or more and make a much larger investment than your normal monthly spend to account for it up front. Once you get really large though you can actually just colocate a bunch of systems and storage and throw a hypervisor manager on them and get almost all the benefits of a cloud based system for an up-front cost that's probably not much more than a dew months of your cloud costs (it looks like 3-4 months is the break-even on some large AWS instances I just looked at compared to some 64-core/512GB boxes I specced out recently). You won't be able to scale quite as immediately to large load increases unless you under-provision, but at that point you're large enough that you should have a handle on what your load spikes are like, and maybe even be able to supplement with could offerings in a pinch. In a way, cloud-based offerings allow medium-to-large businesses to have a sort of insurance on scaling, in that they are paying more for it, but they can stop at the right amount for their need or even back-off because of problems (or engineering coming up with a better way to handle some load).
mbesto 7y ago
> from being strictly growth focused to being more margin focus And hilariously, the margin gains are not in COGS (e.g. computing) but in reductions/efficiencies in S&M and R&D. Most modern SaaS companies are 80% gross margins. > it seems likely that more will consider buying machines instead of always renting them And the companies buying physical hardware will want to know if the vendor supplying the hardware is going to be around to support the hardware in 5 years when they need to be refreshed. It seems like they are building a product that is anti-trend. Seems odd. EDIT: > Cloud is cheaper. A lot cheaper, especially if using managed services This is not universally true.

jonny_eh 7y ago

Netflix famously moved their infrastructure from their own data centres to AWS.

twoodfin 7y ago
Though isn’t that the “commodity” part of their infrastructure? They’re presumably still spending like gangbusters on their “on premises” ISP-local hardware & software. I imagine that’s the extreme end of the market Oxide is targeting: Provide a HW/SW platform to build your differentiation on, because the capabilities and value (& moat) aren’t there in the commodity cloud.
- hinkley 7y ago
  I really think this shouldn't be an either-or thing, most especially for established companies, but to some extent for growing ones as well. If you run a data center in the same city/cities as your engineering strengths are located, that doesn't necessarily give you the geographical diversity you need for many companies. But if you don't run any data centers, you're blind to some of the cost structures, and architectural limitations of the system, and you can become soft. I was going to make a simile to Google and Facebook having their own hardware divisions, but I don't really need to, because they are planning to make cloud-friendly custom hardware for data centers. This seems like an area that the incumbents should be all over but I can't recall the last time I read of innovations from them. Which means there is space for someone new to establish a toe-hold. If I were based in Chicago I'd want a data center in Chicago, and Cloud Servers on the West Coast, (and Europe, etc as applicable). But the lock-in situation is untenable to me right now. It's pretty easy to end up deploying multiple solutions for the same situation. That just complicates reasoning about the system. It bears a resemblance to the Lavaflow antipattern and I can tell you that either can be no fun at all.
  - hinkley 7y ago
    I'm not sure why I chose the phrase 'planning to' about custom hardware, when we know that they've been doing this for a very long time, and FB in particular has not only published designs but multiple cycles of amendments to those designs.
outside1234 7y ago
They also have the size and early mover status where they probably inked a pretty sweet deal to do that.

StacyRoberts 7y ago

The expensive part is the capacity per dollar. It's not cloud vs on prem. The real value play is colocation. Imagine operational costs at $100/mo for a terabyte of RAM and 32 cores on unmetered bandwidth compared to the equivalent at aws. Most sv startups don't care about operational costs but they do care about capital costs of buying the hardware. Primarily due to the very short term focus. I think their own resumes are important too vs their employers' long term viability so everyone pushes aws or azure etc because it's hot tech. Thinking long term is not a silicon valley thing. It's a totally different mindset.

kbenson 7y ago
At the very lowest end, colocation is a clear winner in value, even up to a few servers, it stands up well. Once you have multiple servers, and need to start load-balancing some sets of servers to deal with load spikes, cloud-based infrastructure starts to have some benefits that make it competitive and possible cheaper in some instances, because of load scaling and ease of turn-up of new systems. If you're increasing load by 20% or more a month, you likely will have major problems provisioning servers fast enough unless you gamble that trend will increase for a year or more and make a much larger investment than your normal monthly spend to account for it up front. Once you get really large though you can actually just colocate a bunch of systems and storage and throw a hypervisor manager on them and get almost all the benefits of a cloud based system for an up-front cost that's probably not much more than a dew months of your cloud costs (it looks like 3-4 months is the break-even on some large AWS instances I just looked at compared to some 64-core/512GB boxes I specced out recently). You won't be able to scale quite as immediately to large load increases unless you under-provision, but at that point you're large enough that you should have a handle on what your load spikes are like, and maybe even be able to supplement with could offerings in a pinch. In a way, cloud-based offerings allow medium-to-large businesses to have a sort of insurance on scaling, in that they are paying more for it, but they can stop at the right amount for their need or even back-off because of problems (or engineering coming up with a better way to handle some load).

mbesto 7y ago

> from being strictly growth focused to being more margin focus And hilariously, the margin gains are not in COGS (e.g. computing) but in reductions/efficiencies in S&M and R&D. Most modern SaaS companies are 80% gross margins. > it seems likely that more will consider buying machines instead of always renting them And the companies buying physical hardware will want to know if the vendor supplying the hardware is going to be around to support the hardware in 5 years when they need to be refreshed. It seems like they are building a product that is anti-trend. Seems odd. EDIT: > Cloud is cheaper. A lot cheaper, especially if using managed services This is not universally true.