3 ms·

SQL databases in general don't scale well once they've maxed out a single machine, cross server joins are never going to be pretty and there's a pretty good rea

SQL databases in general don't scale well once they've maxed out a single machine, cross server joins are never going to be pretty and there's a pretty good reason why denormalising data and the adoption of NoSQL solutions with baked in clustering is taking off in a big way. Re NoSQL solutions on windows, theoretically possible, sure, but which NoSQL solution targets windows as its primary platform rather than an afterthought? If you file a bug report, will they notice or care?

Retric 13y ago
In our tests NoSQL was far slower than SQL solutions the advantage was it dealt with hardware failures more easily, but if you use 5% as much hardware it's far less of an issue. Edit: I think one of the largest probmems is most developers doin't understand SQL or databases that well. Indexed Views for example are an incredibly efficient tools that allow you to avoid a lot of expensive joins without the normal risks of denormalizing your data. However, try and think of the last time a developer mentioned them.
- jf22 13y ago
  Sure, but this is a tech agnostic statement. Nothing to do with Microsoft or .net.
- seanc722 13y ago
  Have any resources that explain Views and the efficiency vs a query..? I thought views had the same performance of the underlying SQL statements.
  - Retric 13y ago
    In SQL server normal Views are basically just a select statent but indexed views can have large performance advantages. http://technet.microsoft.com/en-us/library/cc917715.aspx http://technet.microsoft.com/en-us/library/cc917715.aspx Other DB's have there own optimizations for the same idea.
- etherael 13y ago
  Last I saw cassandra and pgsql were pretty close to even on single systems, but you can push as many shards in a cassandra cluster as you want. Where are you getting 5% from? Do you mean to imply sql is typically 20x faster than NoSQL on comparable hardware prior to clustering?
  - Retric 13y ago
    They provide similar performance where there features overlap but Cassandra is still missing important RDBMS optimizations. A great example being transactions, if you actually need that type of protection it takes a lot more overhead to fake if you don't have DB support. That said I am a believer in mixing both approaches it takes a little more thought and effort, but the benefits can be dramatic.
    - etherael 13y ago
      I agree with you here, NoSQL can't be used for everything yet, maybe one day, hyperdex + acid looks promising, we'll see. But even restricting the field back to first tier SQL based solutions, you're still talking MS SQL vs PostgreSQL, and the associated licensing fees involved in both scenarios (expensive vs free). I think performance is pretty similar between the two also last I checked. And you can setup PGSQL with a drbd failover without that much hassle. What are the licensing implications of running two windows servers with an MS SQL server failover? Do you actually need to purchase two copies for that?
jf22 13y ago
Your first sentence is has 0% to do with the tech stack. RavenDb. Yes: https://jira.mongodb.org/browse/SERVER-2612 https://jira.mongodb.org/browse/SERVER-2612 http://www.windowsazure.com/en-us/develop/nodejs/ http://www.windowsazure.com/en-us/develop/nodejs/
- etherael 13y ago
  > Your first sentence is has 0% to do with the tech stack. It does if the target area is massively underrepresented in the tech stack in question. It's just a plain old fact that there are far more options for NoSQL storage layers built with linux first in mind than there are windows. Just like there are far more options for games toolkits and libraries built with windows first in mind than there are for linux. RavenDB I had never even heard of before, but it does look like it fits the bill of NoSQL datastore that primary targets windows, compared to the competition though that doesn't come near to nullifying my point that it's a vastly outscaled matchup. Also, if you even want clustering in RavenDB, you need an enterprise license. There are many completely free NoSQL datastores for linux that handle clustering just fine without an enterprise license. Re MongoDB, it's good that they fixed that, however; https://jira.mongodb.org/secure/IssueNavigator.jspa?sorter/field=resolution&sorter/order=ASC https://jira.mongodb.org/secure/IssueNavigator.jspa?sorter/f... So there are 117 Windows specific issues with mongodb, many of which have not been touched for years or months, and none of which affect linux. I couldn't actually find any tickets which only pertained to linux, though maybe they're there they're certainly not as common.

Retric 13y ago

In our tests NoSQL was far slower than SQL solutions the advantage was it dealt with hardware failures more easily, but if you use 5% as much hardware it's far less of an issue. Edit: I think one of the largest probmems is most developers doin't understand SQL or databases that well. Indexed Views for example are an incredibly efficient tools that allow you to avoid a lot of expensive joins without the normal risks of denormalizing your data. However, try and think of the last time a developer mentioned them.

jf22 13y ago
Sure, but this is a tech agnostic statement. Nothing to do with Microsoft or .net.
seanc722 13y ago
Have any resources that explain Views and the efficiency vs a query..? I thought views had the same performance of the underlying SQL statements.
- Retric 13y ago
  In SQL server normal Views are basically just a select statent but indexed views can have large performance advantages. http://technet.microsoft.com/en-us/library/cc917715.aspx http://technet.microsoft.com/en-us/library/cc917715.aspx Other DB's have there own optimizations for the same idea.
etherael 13y ago
Last I saw cassandra and pgsql were pretty close to even on single systems, but you can push as many shards in a cassandra cluster as you want. Where are you getting 5% from? Do you mean to imply sql is typically 20x faster than NoSQL on comparable hardware prior to clustering?
- Retric 13y ago
  They provide similar performance where there features overlap but Cassandra is still missing important RDBMS optimizations. A great example being transactions, if you actually need that type of protection it takes a lot more overhead to fake if you don't have DB support. That said I am a believer in mixing both approaches it takes a little more thought and effort, but the benefits can be dramatic.
  - etherael 13y ago
    I agree with you here, NoSQL can't be used for everything yet, maybe one day, hyperdex + acid looks promising, we'll see. But even restricting the field back to first tier SQL based solutions, you're still talking MS SQL vs PostgreSQL, and the associated licensing fees involved in both scenarios (expensive vs free). I think performance is pretty similar between the two also last I checked. And you can setup PGSQL with a drbd failover without that much hassle. What are the licensing implications of running two windows servers with an MS SQL server failover? Do you actually need to purchase two copies for that?

jf22 13y ago

Your first sentence is has 0% to do with the tech stack. RavenDb. Yes: https://jira.mongodb.org/browse/SERVER-2612 https://jira.mongodb.org/browse/SERVER-2612 http://www.windowsazure.com/en-us/develop/nodejs/ http://www.windowsazure.com/en-us/develop/nodejs/

etherael 13y ago
> Your first sentence is has 0% to do with the tech stack. It does if the target area is massively underrepresented in the tech stack in question. It's just a plain old fact that there are far more options for NoSQL storage layers built with linux first in mind than there are windows. Just like there are far more options for games toolkits and libraries built with windows first in mind than there are for linux. RavenDB I had never even heard of before, but it does look like it fits the bill of NoSQL datastore that primary targets windows, compared to the competition though that doesn't come near to nullifying my point that it's a vastly outscaled matchup. Also, if you even want clustering in RavenDB, you need an enterprise license. There are many completely free NoSQL datastores for linux that handle clustering just fine without an enterprise license. Re MongoDB, it's good that they fixed that, however; https://jira.mongodb.org/secure/IssueNavigator.jspa?sorter/field=resolution&sorter/order=ASC https://jira.mongodb.org/secure/IssueNavigator.jspa?sorter/f... So there are 117 Windows specific issues with mongodb, many of which have not been touched for years or months, and none of which affect linux. I couldn't actually find any tickets which only pertained to linux, though maybe they're there they're certainly not as common.