Why should I use snapshot shipping rather than online (DRBD Proxy) with LINBIT SDS?
The are a couple of reasons why we direct customers to continually repeated (scheduled) snapshot delta shipping in LINSTOR:- Repeated shipping of snapshot deltas can handle widely varying network bandwidth.
For example, assume that the network link between the two sites that you are shipping snapshot deltas between is shared. A shared link could mean shared with different services or applications running in the two sites, a shared link with other customers, or even a network path over the internet. When the link between your two sites is not a dedicated link with guaranteed exclusive bandwidth, the available network bandwidth might be lower than the bandwidth required to replicate the writes of the Kubernetes cluster's applications.
Those bandwidth differences will fill DRBD Proxy's buffer. When the buffer runs full, the application on the source side will be slowed down.
In contrast, with repeated snapshot delta shipping, you will never experience a slowdown of applications due to limited bandwidth on the site link. Instead, the system might decrease the frequency of snapshot delta shipping in response to the limited bandwidth. - Repeated snapshot delta shipping is more efficient regarding overwriting writes.
Imaging a database that frequently updates a WAL ring buffer area on the disk. With DRBD's real-time replication, all these overwrites get replicated as they are made. However, with snapshot delta shipping, only the data of the final overwrite in each snapshot shipping interval gets replicated. - Repeated shipping of snapshot deltas could be configured to happen every five minutes. That gives you a defined RPO.
Please note: While the concept is repeatedly shipping snapshot differences, this functionality is called "backup shipping" in LINSTOR. It is documented in the LINSTOR users guide.
What are the advantages of DRBD-Proxy, then?
- With DRBD and DRBD-Proxy, you do online data replication. That means every write operation to the data volume is mirrored to the other site asynchronously. Your RPO is nearly zero when the network bandwidth is higher than your write bandwidth.
- It is a good fit when you have a dedicated (=not shared) network link for your site-to-site replication.
The LINSTOR users guide contains a chapter on configuring DBRD-Proxy. It might change in the future to integrate better with resource groups.
Created 2022/12/13 – PR
Reviewed 2022/12/14 – DV, MAT