A comparison of Scale-up vs Scale-out Replication Technologies
By Joshua Stenhouse, Zerto Technical Evangelist
When evaluating which hypervisor-based continuous replication and recovery solution to use for your virtual environment, an increasingly important consideration is whether to use a scale-up or scale-out hypervisor based replication technology. These are defined as:
- Scale-up replication technologies rely on pre-sized virtual appliances that are often fixed in size, large in footprint, and operate in a single or redundant architecture that replicate VM block-level changes from multiple hypervisor hosts requiring dedicated network connectivity for write traffic
- Scale-out replication technologies rely on virtual appliances with a small footprint, that don’t require pre-sizing, that are deployed one per hypervisor host to replicate only the VM block-level changes from that host removing the need for dedicated network connectivity
When considering which replication technology to use there are some extremely important considerations to take into account.
Pre-write or Post-write Replication Engines
Replicating pre-write means that all protected VM writes have to be processed and confirmed by the replication appliance before they are written to the source storage, introducing latency and limiting the maximum IOPs and throughput of all protected VMs to the speed of the replication appliance.
Scale-up replication technologies use pre-write replication to confirm receipt of the changed blocks from multiple hosts which then has to been written to a local journal before the write can proceed. This permanently impacts production performance as the protected VMs can only ever write as fast as the replication appliances and local journal allow, typically resulting in a 5x reduction in IOPs, throughput and 5x increase in latency on all protected VM writes with all production writes doubled in the source storage.
Replicating post-write means the writes are replicated after being written to the source storage and therefore no latency, IOPs or throughput performance degradation is added to the protected VMs. Scale-out technologies utilize post-write replication to allow the VMs to utilize the maximum performance of the storage they are stored on and not be limited by the performance of the replication appliance with no overhead on the source storage array.
Scale-out and post-write replication
Write Traffic Networking
Scale-up replication technologies require a dedicated iSCSI network between each hypervisor host and replication appliance for the hosts to send the changed blocks for replication. This introduces significant production changes as physical network adapters, dedicated vSwitches, port groups and VLANs are required. This also means that all the benefits of using fibre attached storage are negated for VM writes as they are sent over an iSCSI network first.
With scale-out replication all of the protected VM changed blocks are sent in the memory of the hypervisor host with no performance overhead or degradation. No major production changes are needed and no dedicated iSCSI network for protected VM writes is needed.
Sizing and Scalability
Due to the performance characteristics of scale-up replication technologies, pre-sizing is often one of the most important aspects to try minimize the performance impact while anticipating future growth and scalability. Failure to correctly size the appliances can result in severely degrading the protected VM performance. Scale-out replication require no detailed sizing exercises as there is no performance impact and the power of the replication engine increases with the size of the environment in line with the number of hypervisor hosts.
With scale-up replication technologies it is recommend to keep protected VMs on a separate host to the active replication appliance. As the replication appliances have to be deployed in pairs, on separate hosts for redundancy, this means that 2 dedicated ESXi hosts have to be kept separate from protected VMs for best practice in the event of a HA failover between appliances. This incurs a significant production hardware and hypervisor license cost overhead.
Scale-out replication technologies utilize one appliance per host, removing the need for dedicated hosts, reducing the overhead of the replication appliances on expensive infrastructure resources.
Zerto Virtual Replication is the only scale-out hypervisor-based replication solution designed for the hypervisor that scales in-line with the protected VM environment by deploying one appliance per host. No dedicated iSCSI networking or hosts are required and no performance penalty is introduced to the protected VMs.
With advanced features such as bandwidth throttling, multi-tenancy, vCD integration, traffic prioritization, cross-hypervisor replication, SQL and Oracle TempDB optimization and flexible journaling it is the only truly scalable hypervisor based replication solution available in the market today.