DO Ideas 2

Amazon EBS like storage

I would be great if EBS like storage option is available. All our data can be stored in a separate drive instead of having everything in the OS partition. How many votes for this ?

  • Arul Kumar
  • Sep 11 2018
  • Will not implement
  • Attach files
  • Anonymous commented
    September 11, 2018 19:01

    Erm, fuse? If you don't care about the performance hit of NAS then you can use ... Though Difital Ocean desperately needs solutions with more disk space.

  • NP commented
    September 11, 2018 19:01

    You guys will never get the type of client base like Rackspace, Softlayer, Amazon, etc... without a solution that provides storage addition to the VM guests. Fails to meet my needs, and I am just someone who deals with only about 10 companies. Rethink this.

  • Moisey Uretsky commented
    September 11, 2018 19:01

    Unfortunately we will not be implementing an EBS solution because storing virtual server's primary drives on a central storage system creates large performance issues.

    While it does provide benefits like easy failover the issue is that a failed virtual server should occur about 1% of the time, which means 99% of the time that you are on EBS you aren't getting any benefit from it and instead are suffering from worse performance as a result.

    We decided to use local storage to avoid this large performance bottleneck for an issue which can be easily resolved with todays modern programming frameworks and open source projects, such as using nginx as a load balancer with multiple web servers.

    Additionally by using a large centralized storage system a single point of failure is created that can at once affect thousands of customers. In most cases when these failures occur they are also fairly complex and its not as simple as "reboot" it to get things back up and running.

    If you review the multiple failures that AWS has recently suffered the majority of them are tied to EBS, compare that to a simple distributed "dumb" system like S3 which suffers almost no failures ever.

    Certainly the workloads are different, but it shows that piecing together simpler building blocks which are independent while more time consuming at the beginning yields a better overall continuity of service.