Veeam Home | Support | Downloads
Best Practices for Veeam Backup & Replication 7.0 for VMware

Veeam Management Pack Help Center  > Best Practices for Deployment & Configuration >  Resource Planning and Optimization > Planning for Proxies > Sizing


Previous page Next page Print this Topic

Table of contents

As per system requirements, a proxy requires 2 GB RAM plus 200MB for each concurrent task (VM disk to be processed).

Determining the exact number of proxies that is ideal for a specific environment can difficult, but the following recommendations are provided as starting points. Significant variation can occur based on many factors, including bottlenecks in other areas such as source/target storage performance, or network bandwidth. For example, if the source storage is able to provide 400MB/s of throughput, and two proxies are each doing 200MB/s, then adding an additional proxy is unlikely to improve backup performance since they are already at the source storage capacity. It is also recommended to deploy additional resources to handle the load in case of failure of a proxy, deploying N+1 proxies to provide redundancy. The following estimations can be of use:

Virtual Proxy — 1 vCPU of proxy capacity for every 200 VMs, at least one proxy per cluster

Physical Proxy — 1 CPU core of proxy capacity for every 250VMs


Due to the overhead of virtualizing SMP system with high vCPU counts, virtual proxies generally provide peak efficiency at 4vCPUs. In most environments better performance is achieved by deploying two 4vCPU proxies on two different hosts rather than a single 8vCPU proxy.

These guidelines assume industry typical environments, with average VM sizes of 50-100GB, daily change rates of ~5%, and an 8 hour backup window (initial full backup may take longer). Note that there is also an assumption of high-speed interconnects such as 8Gb Fiber Channel or 10Gb Ethernet. Environments that leverage lower speeds connectivity such a 1Gb iSCSI, NFS, or SMB/CIFS targets may need more, smaller proxies to take advantages of multiple links.


Deploying test proxies and running jobs to measure throughput can help determine more accurate numbers for the specific environment.

Table of contents