DR and multi-data center HA architectures


Portworx offers multiple options for DR and multi-data center HA beyond what is provided with the single data center/multiple AZ deployment options. For more on why you cannot simply use your traditional DR system for Kubernetes applications, see the Limits of traditional DR for Kubernetes applications section.

Option 1- Synchronous DR over a MAN using multiple Kubernetes clusters with a single Portworx stretch cluster

Solves for these primary needs:

  • Container- or Namespace-granular DR.
  • RPO zero data protection and low RTO in the case of losing an entire data center.
  • Data and Kubernetes objects are replicated across both sites, simplifying and speeding up application failover

Key constraints:

  • Requires two data centers.
  • Round trip latency between data centers must be < 10 ms. This can usually be accomplished using different cloud providers in the same geographical region, colocating an on-premises data center in the same region as a cloud provider or operating a campus or metropolitan area network.

Synchronous Disaster Recovery over a metro area network using multiple Kubernetes clusters with a single Portworx stretch cluster

Option 2- Asynchronous DR over a WAN using multiple Kubernetes clusters with multiple Portworx clusters

Solves for these primary needs:

  • Container- or Namespace-granular Disaster Recovery.
  • RPO levels are as low as 15 minutes and RTO under one minute in the case of losing an entire data center.
  • Data and Kubernetes objects are replicated across both sites, simplifying and speeding up application failover

Key constraints:

  • Requires two data centers.
  • S3-compatible object store is available to move data.

Asynchronous DR of a wide-area network (WAN) using multiple Kubernetes clusters with multiple Portworx clusters

Summary of Synchronous vs. Asynchronous PX-DR Options

Application and infrastructure requirements Synchronous PX-DR Asynchronous PD-DR
Number of Portworx clusters 1 2
Needs an S3-compatible object store to move data No Yes
Max round trip latency between data centers 10 ms No limit
Data guaranteed to be available at both sites (zero RPO) Yes No
Kubernetes objects replicated between data centers Yes Yes
Low RTO Yes Yes

Option 3- Multi-site data center stretch cluster for HA

Solves for these primary needs:

  • High availability of data services across multiple data centers.
  • Zero RPO data protection.

Key constraints:

  • Requires three data centers.
  • Max round trip latency between data centers of under 10 ms.

Multi-site data center stretch cluster for HA


Last edited: Friday, Aug 7, 2020