5 ms·

I don't see what the anecdote about the other person has to do with your story. It seems like you're citing it to try to imply that Linux can be expensive, but

by ebbv 13y ago

I don't see what the anecdote about the other person has to do with your story. It seems like you're citing it to try to imply that Linux can be expensive, but that's not really a valid point based on that anecdote. Their software costs were $0/mo of that $3000/mo. Linux has nothing to do with whatever reasons they had for such expensive server costs. There is no question that whatever their server costs, they would have been even higher if they had been Microsoft based. Secondly, I'd disagree with your assertion that the customer doesn't matter. Customers may not care when they're using your SaaS product, but my company is in the market for new support software and we looked at yours but because yours is Microsoft based, it's disqualified. We are looking for a self hosted solution. So you lost at least one customer right here. I'm not making this up, this all happened a couple weeks ago prior to this blog post. EDIT: PS - I liked your previous blog post about A/B testing, so it's nothing personal I just think this one is kinda bogus.

codeulike 13y ago
Linux has nothing to do with whatever reasons they had for such expensive server costs. It might do. The point is that Node/Ruby etc aren't particularly efficient compared to C# and the CLR, so you need more Server power for a given application. For example StackOverflow ran from a single windows server in the early days, they didn't add a second one till Oct 2008 http://blog.stackoverflow.com/2008/10/adde-a-second-server/ http://blog.stackoverflow.com/2008/10/adde-a-second-server/
- KaiserPro 13y ago
  A bad[1] workman is perfectly capable of being inefficient with any tool. [1]this isn't passing comment on any of the people referenced in the original post
- spamizbad 13y ago
  Techempower's benchmark's paint a different story than you describe. Node.js seems to be the fastest. ASP.NET MVC shows some strong performance in a handful of database tests but when it comes things like json serialization etc it falls behind the competition. http://www.techempower.com/benchmarks/#section=data-r5&f=ug-13ydj4-0 http://www.techempower.com/benchmarks/#section=data-r5&f=ug-... Edit: FWIW it doesn't test Microsoft SQL Server. Still, for most simple-ish web applications I can't imagine MSSQL server being significantly faster than Postgres or MySQL. Its selling point is really its feature set as opposed to its performance anyway.
  - TheAnimus 13y ago
    >Edit: FWIW it doesn't test Microsoft SQL Server. Still, for most simple-ish web applications I can't imagine MSSQL server being significantly faster than Postgres or MySQL. Its selling point is really its feature set as opposed to its performance anyway. More than just that, they tested mono, which sadly is often very slow, and didn't test on IIS.
    - bhauer 13y ago
      The ASP.NET tests run on Windows (EC2) used IIS. We have a SQL Server pull request [1] but haven't been able to incorporate it yet and this will not be included in the upcoming Round 6. https://github.com/TechEmpower/FrameworkBenchmarks/pull/336 https://github.com/TechEmpower/FrameworkBenchmarks/pull/336
      - TheAnimus 13y ago
        The problem is the comparisons are a bit tricky. For instance why is the performance in the Fortune example so good, yet so bad in the simple multirequest response. Ultimately, I think people have to bench with something in mind for their intended usages. If someone (as in this thread) is doing some massive data gathering with 6000+ nodes, well windows licenses will be prohibitive against any dev time saving, and ultimately, performance is going to be optimised so heavily a 'thinner' OS is obviously going to be a possible better choice. This isn't always the case, for us the agile nature of what we are doing, the fact it's only a couple of hundread nodes means that we will have different demands. Platform Performance, Platform Cost, Platform Dev Costs. It is just an exercise between those three. Looking at some of the comparisons there, it reminds me of the whole BSD vs Linux thing that was so 2004. Oh BSD takes a few extra op codes to do hello world.... The problem is some people never include the use of say MVC4 or C#. It isn't that they are missing out, it is that they are so vocal about how bad the platform they have never used is. The last HTML site I delivered, I could have done in about a third of the time, with much more reliable user experience if we had used Silverlight. It was an internal project too. Yet someone who had never even used anything but HTML decided it had to be a website. I think people are just trying to rally against that cry. All I was trying to do is point out that those simple number comparisons are not only flawed / unfair but also irrelevant. There isn't a one size fits all technology, there never will be.
- ebbv 13y ago
  I'm not going to argue efficiency of node vs. C# here, it's too complicated and I don't have a good set of benchmarks on hand. But I guarantee you that the traffic that the article describes could be handled by Node and MySQL or MariaDB on a single server too. Just because someone is inefficient with certain tools doesn't mean those tools are themselves inefficient.
- jlarocco 13y ago
  "Node/Ruby are slow, therefore Linux is not good for servers." I don't think that argument works. There are many, many other tools besides Node/Ruby that run on Linux. Many of them are as efficient or more efficient than C# and the CLR.
- Toshio 13y ago
  > "The point is that Node/Ruby etc aren't particularly efficient [...]" Any hard data you can point to in support of this rather bizarre statement?
- dragonwriter 13y ago
  > The point is that Node/Ruby etc aren't particularly efficient compared to C# and the CLR, so you need more Server power for a given application. Node and Ruby aren't Linux (they run on all kinds of platforms, including both Linux and Windows), and Node and Ruby aren't the only alternatives to C# and the CLR (either on Linux or anywhere else).
TheAnimus 13y ago
> Linux has nothing to do with whatever reasons I think it is not about Linux in isolation, but about the fancy, over hyped tech stacks you see people using. About 6 months ago I was having a few days off, out in the wilderness so to speak, I decided to try this whole Go thing. When reading about it, I thought nice I like the actor paradigm, I was doing a prime solver from someone's tutorial, and... It was slow. Not slow, slow, but slow. Compared the same thing to a similar idea in C# in Rx and in TPL, and in F# and in CamelLight, with one in Java. For my preference the syntax of C# was the best, and funnily enough the old fashioned style implementation I did in Java was the fastest, but not by much at all. Go was about 20% of the speed of the java one, not much faster than the CamlLight implementation. This wasn't scientific, this wasn't a fair comparison, but I am cynical about the 'buzz' or hype for some of these frameworks. People often view things like Node.js as free. Which isn't true when you are paying for the hardware it runs on. The 'cost' of C# or MS SQL server isn't just the Microsoft licenses either. You can't look at these things solely in isolation. However its rather un-interesting to say that a behemoth and status quo are actually good options. 'new' things are much more interesting, so get spoken about a lot more, but I'd love to know something truly high scale, highly available in Node.js, and how much it has really cost them.
- throwit1979 13y ago
  Go was about 20% of the speed of the java one You're probably doing something wrong. Go tops the techempower benchmarks for concurrent db access and page rendering, and for pure math performance, it has an average case performance of 17% slower than Java( highly dependent on problem chosen of course ) refs: http://www.techempower.com/benchmarks/ http://www.techempower.com/benchmarks/ http://benchmarksgame.alioth.debian.org/u64q/benchmark.php?test=all&lang=go&lang2=java&data=u64q http://benchmarksgame.alioth.debian.org/u64q/benchmark.php?t...
  - pstuart 13y ago
    And that's with Go 1.1 vs. Java 6 or 7.
    - 13y ago
tommis 13y ago
I'm curious, why would you disqualify a SaaS product solely because it's based on MS?
- ebbv 13y ago
  Read again: > Customers may not care when they're using your SaaS product, but my company is in the market for new support software and we looked at yours but because yours is Microsoft based, it's disqualified. We are looking for a self hosted solution. OP offers a self hosted solution as well.

by ebbv 13y ago

codeulike 13y ago

Linux has nothing to do with whatever reasons they had for such expensive server costs. It might do. The point is that Node/Ruby etc aren't particularly efficient compared to C# and the CLR, so you need more Server power for a given application. For example StackOverflow ran from a single windows server in the early days, they didn't add a second one till Oct 2008 http://blog.stackoverflow.com/2008/10/adde-a-second-server/ http://blog.stackoverflow.com/2008/10/adde-a-second-server/

KaiserPro 13y ago
A bad[1] workman is perfectly capable of being inefficient with any tool. [1]this isn't passing comment on any of the people referenced in the original post
spamizbad 13y ago
Techempower's benchmark's paint a different story than you describe. Node.js seems to be the fastest. ASP.NET MVC shows some strong performance in a handful of database tests but when it comes things like json serialization etc it falls behind the competition. http://www.techempower.com/benchmarks/#section=data-r5&f=ug-13ydj4-0 http://www.techempower.com/benchmarks/#section=data-r5&f=ug-... Edit: FWIW it doesn't test Microsoft SQL Server. Still, for most simple-ish web applications I can't imagine MSSQL server being significantly faster than Postgres or MySQL. Its selling point is really its feature set as opposed to its performance anyway.
- TheAnimus 13y ago
  >Edit: FWIW it doesn't test Microsoft SQL Server. Still, for most simple-ish web applications I can't imagine MSSQL server being significantly faster than Postgres or MySQL. Its selling point is really its feature set as opposed to its performance anyway. More than just that, they tested mono, which sadly is often very slow, and didn't test on IIS.
  - bhauer 13y ago
    The ASP.NET tests run on Windows (EC2) used IIS. We have a SQL Server pull request [1] but haven't been able to incorporate it yet and this will not be included in the upcoming Round 6. https://github.com/TechEmpower/FrameworkBenchmarks/pull/336 https://github.com/TechEmpower/FrameworkBenchmarks/pull/336
    - TheAnimus 13y ago
      The problem is the comparisons are a bit tricky. For instance why is the performance in the Fortune example so good, yet so bad in the simple multirequest response. Ultimately, I think people have to bench with something in mind for their intended usages. If someone (as in this thread) is doing some massive data gathering with 6000+ nodes, well windows licenses will be prohibitive against any dev time saving, and ultimately, performance is going to be optimised so heavily a 'thinner' OS is obviously going to be a possible better choice. This isn't always the case, for us the agile nature of what we are doing, the fact it's only a couple of hundread nodes means that we will have different demands. Platform Performance, Platform Cost, Platform Dev Costs. It is just an exercise between those three. Looking at some of the comparisons there, it reminds me of the whole BSD vs Linux thing that was so 2004. Oh BSD takes a few extra op codes to do hello world.... The problem is some people never include the use of say MVC4 or C#. It isn't that they are missing out, it is that they are so vocal about how bad the platform they have never used is. The last HTML site I delivered, I could have done in about a third of the time, with much more reliable user experience if we had used Silverlight. It was an internal project too. Yet someone who had never even used anything but HTML decided it had to be a website. I think people are just trying to rally against that cry. All I was trying to do is point out that those simple number comparisons are not only flawed / unfair but also irrelevant. There isn't a one size fits all technology, there never will be.
ebbv 13y ago
I'm not going to argue efficiency of node vs. C# here, it's too complicated and I don't have a good set of benchmarks on hand. But I guarantee you that the traffic that the article describes could be handled by Node and MySQL or MariaDB on a single server too. Just because someone is inefficient with certain tools doesn't mean those tools are themselves inefficient.
jlarocco 13y ago
"Node/Ruby are slow, therefore Linux is not good for servers." I don't think that argument works. There are many, many other tools besides Node/Ruby that run on Linux. Many of them are as efficient or more efficient than C# and the CLR.
Toshio 13y ago
> "The point is that Node/Ruby etc aren't particularly efficient [...]" Any hard data you can point to in support of this rather bizarre statement?
dragonwriter 13y ago
> The point is that Node/Ruby etc aren't particularly efficient compared to C# and the CLR, so you need more Server power for a given application. Node and Ruby aren't Linux (they run on all kinds of platforms, including both Linux and Windows), and Node and Ruby aren't the only alternatives to C# and the CLR (either on Linux or anywhere else).

TheAnimus 13y ago

> Linux has nothing to do with whatever reasons I think it is not about Linux in isolation, but about the fancy, over hyped tech stacks you see people using. About 6 months ago I was having a few days off, out in the wilderness so to speak, I decided to try this whole Go thing. When reading about it, I thought nice I like the actor paradigm, I was doing a prime solver from someone's tutorial, and... It was slow. Not slow, slow, but slow. Compared the same thing to a similar idea in C# in Rx and in TPL, and in F# and in CamelLight, with one in Java. For my preference the syntax of C# was the best, and funnily enough the old fashioned style implementation I did in Java was the fastest, but not by much at all. Go was about 20% of the speed of the java one, not much faster than the CamlLight implementation. This wasn't scientific, this wasn't a fair comparison, but I am cynical about the 'buzz' or hype for some of these frameworks. People often view things like Node.js as free. Which isn't true when you are paying for the hardware it runs on. The 'cost' of C# or MS SQL server isn't just the Microsoft licenses either. You can't look at these things solely in isolation. However its rather un-interesting to say that a behemoth and status quo are actually good options. 'new' things are much more interesting, so get spoken about a lot more, but I'd love to know something truly high scale, highly available in Node.js, and how much it has really cost them.

throwit1979 13y ago
Go was about 20% of the speed of the java one You're probably doing something wrong. Go tops the techempower benchmarks for concurrent db access and page rendering, and for pure math performance, it has an average case performance of 17% slower than Java( highly dependent on problem chosen of course ) refs: http://www.techempower.com/benchmarks/ http://www.techempower.com/benchmarks/ http://benchmarksgame.alioth.debian.org/u64q/benchmark.php?test=all&lang=go&lang2=java&data=u64q http://benchmarksgame.alioth.debian.org/u64q/benchmark.php?t...
- pstuart 13y ago
  And that's with Go 1.1 vs. Java 6 or 7.
  - 13y ago

tommis 13y ago

I'm curious, why would you disqualify a SaaS product solely because it's based on MS?

ebbv 13y ago
Read again: > Customers may not care when they're using your SaaS product, but my company is in the market for new support software and we looked at yours but because yours is Microsoft based, it's disqualified. We are looking for a self hosted solution. OP offers a self hosted solution as well.