Migrating a large site from SQL Server to MongoDB - Lessons Learned

[rfurlan]

SQL and NoSQL are different tools for different jobs. A couple of months ago we migrated a large site from SQL Server to MongoDB, it was a bumpy road and I decided to write this article to share what we learned: http://www.wireclub.com/development/TqnkQwQ8CxUYTVT90/read

[amrith]

I'm shocked and mortified ... Check out http://bit.ly/v1cznd

Why would you conclude that MongoDB was "fast" and SQL Server was slow? Did you run SQL server on the same SSD's and multiple HDD's?

to willvarfar's comments, what was the real fsync cost and didn't you get ACID with SQL Server? Now, if ACID is not really required, and you really really need flexible schemas, then by all means ... I understand that you want distributed-queryable-notepad, not a relational database!

[rfurlan → amrith]

Also, I never said SQL Server is slow - it isn't, anyone that would say otherwise must have an agenda, which is not my case. SQL Server is an incredible database and it has served us well.

[rfurlan → amrith]

MongoDB has the potential to be faster than traditional relational databases simply because it does less, way less. The trade off makes sense for some applications, but not for all, it really depends on your requirements.

[rfurlan]

I also hosted a Q&A about our experience here on HN: http://news.ycombinator.com/item?id=3141331

[byoung2]

Clickable: http://www.wireclub.com/development/TqnkQwQ8CxUYTVT90/read

[rfurlan → byoung2]

Thanks byoung2!

[willvarfar]

So do you have fsync enabled?

[rfurlan → willvarfar]

fsync is enabled for the vast majority of calls and it is our default setting. Disabling fsync should be treated as case-by-case optimization IMO

[latch → rfurlan]

I assume when people talk about fsync, they actually mean journaling?

I'm not sure if there's ever a good reason to use {fsync: true} vs {j: true}

[rfurlan → latch]

Fsync means "wait for acknowledgement after each request", if Fsync is off, all calls are fire-and-forget, no errors are reported. There are a few Fsync levels though, you can use it to make the call block until the data has been relayed to at least N other nodes.

[latch → rfurlan]

no, fsync means wait for the data file to be synced to disk. {safe:true} is what you are talking about. {j:true} is probably what yo should be using, {w:majority} is also important, and {fsync:true} is the data file option

I think given that one of the parameters is called "fsync" using "fsync" should mean that.

[rfurlan → latch]

Yes latch, you are correct, thank you