JetStream DR software uses Azure Blob storage as a target storage site to deliver its disaster recovery service. The recommended level of Azure Blob storage service is:
- Standard (GPv2) storage: For general purpose access to block blob and Azure Storage features.
- Hot access tier: Optimized for storing data that is accessed frequently.
- Default configuration settings are used.
In operation, JetStream DR creates a container for each designated group of protected VMs known as “Protected Domains.” Protected Domains store their VMs’ configuration and protected data in a compressed format as block blobs in a simple directory hierarchy (using ‘/’ as the delimiter). JetStream DR does not use hierarchical name space which is more expensive to operate and does not provide additional benefit to the software.
Access
JetStream DR performs operations through APIs using the Azure Storage Client Library for C++ SDK and the Azure Java SDK:
- Container: create/list/delete/exists
- Block blob: put/get/delete/exists
- List blobs in flat/hierarchical structure with prefix filter parameter
JetStream software uses the following blob access patterns:
- Putting objects into the Container (multi-streaming mode)
- Getting objects sequentially from the Container (multi-streaming mode)
- Concurrent putting and getting the above mentioned objects
- Deleting objects randomly
- In rare cases, getting objects randomly (used for “point-in-time” recovery only)
Performance
Successful operation of JetStream DR generally requires high bandwidth and latency to fall within a “reasonable spectrum.” Although there is no hard and fast rule about latency, it was noted in one extreme case an installation site’s latency was measured being up to 5 minutes! Obviously, such high latency would negatively impact all aspects of DR and operation of other software programs.
Optimizing the following will ensure efficient operation of JetStream DR software:
- Memory: Data recovery operations can greatly benefit if system memory and other resources are adequately provisioned. This is an important area to monitor and correctly size.
- Bandwidth: The most important factor that influences recovery SLA (i.e., RTO) is supporting high sequential (multi-streaming) getting objects. Having adequate bandwidth is crucial for successful operation.