Blog

Delphix 4.0 -Faster, Better (III)

I posted a series of blogs talking about performance improvements to some key workflows (SnapSync and

I posted a series of blogs talking about performance improvements to some key workflows (SnapSync and LogSync) in Delphix 4.0. This is the concluding post in that series. In this post I will talk about performance improvements we made to another workflow, Virtual To Physical and also touch on some of the improvements to our filesystem, DxFS.

V2P (Virtual To Physical)

V2P or Virtual to Physical is a unique feature that allows customers to 'export' a database out of Delphix. V2P allows customers to bring up a physical database from any point in the timeflow of a dSource or a VDB.

Since Delphix stays in sync with the production database and delivers non-production app/dev/test environments, V2P is intended to become part of a continuous data protection strategy for both the production as well as the non-production environments. V2P is typically faster compared to loading a DB from a full backup -provides superior RTO (Recovery Time Objective). In addition to data protection, V2P is also useful for upgrade testing.

v2p graphic

Customers can test their upgrades on a VDB and then V2P out to the upgraded physical production database. Some of our customers have processes that dictate that physical copies must be present in remote location for compliance. In such instances, V2P is used to create instances from specific points in the time flow. All these use cases dictate time constraints around V2P.

Before 4.0, V2P used SCP as the underlying transport and was throttled at a maximum throughput of 25 MB/sec due to limitations in the ssh protocol. In 4.0, we updated V2P to use DSP as the underlying transport. Using DSP, V2P throughput increased 4x on a 1GB/sec link. For some of the critical use cases for V2P, this gives back valuable time to the customer.

Customers can now optionally enable compression (to get better utilization on slow links) and encryption when data is sent over in-secure links. The chart above shows V2P throughput (under network bandwidth limited conditions) in Delphix 3.2 compared to different options in 4.0. Note that the native encryption in DSP fares slightly better compared to SCP.

DxFS

All the features I talked about till now add configurability to workflows that are critical to our customers. We want our customers to use Delphix like a Ferrari or a Prius, whatever meets their needs. But when it comes to DxFS, the filesystem which forms the underpinnings of Delphix, the gloves are off.

We want to punch out as much performance from the underlying infrastructure and run as fast as the storage would allow. Down here, we want it to be a Ferarri, all the time. Performance improvements in the filesystem will impact a wide range of workflows that are relevant to our customers. My colleague George made significant strides in improving the filesystem performance.  

As a result of these changes to DxFS, we see an almost 30% increase in random write performance - an especially difficult to handle workload. George is working on a post with more details on the changes made to DxFS, please stay tuned for that.

Conclusion

As we build our Agile Data platform and expand into newer use cases, we see Delphix as the primary vehicle for delivering data to our customers' application projects. Improvements to SnapSync and LogSync will help keep Delphix in sync with production databases efficiently. Depending on the use case, these can be configured to match the activity levels of production databases, while exerting minimal load on the infrastructure.

V2P has been enhanced to deliver almost 4x increase in throughput. The enhancements to DxFS enable us to extract more performance from the storage provisioned to Delphix. With all these features in Dune, we built Delphix to deploy seamlessly and adapt to our customers' environments.