FSx for OpenZFS: compute and network scalability (and fault-tolerance)
How do I address the compute and network scalability of a ...
- read-write,
- dynamically growing, and
- arbitrarily large
file, sitting on a FSx for OpenZFS file-system?
Secondly, since hardware fault-tolerance is usually achieved via an active-active or active-passive cluster, how do I achieve such fault-tolerance with FSx for OpenZFS?
Context: After observing less than desirable latency numbers in Kafka, we're being forced to think implementing our own low-latency subset of Kafka / Redis (with their select features only, and not their complete rewrite!) for our needs. For this implementation, we intend to use OpenZFS for its excellent storage scalability, but we aren't so sure what sort of concurrent user access (read-write) FSx for OpenZFS would support.
- 최신
- 최다 투표
- 가장 많은 댓글
To address the scalability and fault tolerance of a large, dynamically growing file on FSx for OpenZFS:
- FSx for OpenZFS provides fully managed and elastic storage that can scale to petabytes in size. The file system performance also scales automatically as more storage is added
- For fault tolerance, FSx replicates your data across different physical storage infrastructure within an Availability Zone. It also continuously monitors for hardware failures and will automatically replace failed components.
- You can take snapshots of the file system to backup your data and enable point-in-time recovery. Snapshots can be shared across accounts and regions for disaster recovery.
- For higher availability, you can use a multi-AZ deployment of FSx across two Availability Zones. This provides protection from an AZ failure.
- The OpenZFS file system itself provides features like data compression, checksumming, and self-healing that help ensure the integrity and durability of very large files.
@Giovanni The OpenZFS file-system will be NFS-mounted on each VM/container in an elastic pool of VMs/containers. But will the NFS server, providing this export from the OpenZFS side, be able to elastically scale itself somehow and handle read-write requests from millions of users concurrently hitting these VMs/containers?
In other words, while I'm sure that OpenZFS will scale storage capacity wise, I'm not sure if it will also auto-scale with compute and network loads as well.