Disk backup systems (DXi, DD, etc) performance figures

[raphaelmsx]

Hi, I'm evaluating disk backup system for the company's data center.

I currently have the following scenario:

4 VMware ESXi 5.1 hosts and vCenter 5.1 with 8TB of VMs on two IBM DS3524 8Gb FC SAN with 48 600GB 10K disks, one physical server running Windows 2008 and CA ARCserve backing up to a 8Gb FC LTO-5 Tape Library.

I also have on the SAN two IBM DS3512 with twelve 1TB 7.2K NL-SAS which are unused at the moment.

The backup application will be migrated to Veeam, so I already tested backing up to the DS3512 with Veeam, but the throughput was very slow, just 57MB/sec wich translates to only 0.20TB/hour, and this using RAID10 on the DS3512.

The proxy is a VM, not the physycal server that Veeam is installed, and the bottlneck was 99% Source followed by 69% Proxy, so what could be done to improve this performance?

Anyway, looking at the specs of those disk backup devices, I can't really understand how it can achieve such very high throughput.

For example, in the specs of the Quantum DXi6702, it lists a performance of 6.5TB/hour on the NAS interface, but how it can achieve
such throughput, does it need to use the two buil-in 10GbE ports with link aggregation, which would give a theoretical throughput of about 8.78TB/hour?

And looking at the DXi6701 specs, it shows a performance of 9.1TB/hour on the two 8Gb FC ports, I presume that two 8Gb FC ports equals to only 7TB/hour.

Also, I presume that I will need the source VMs on very high speed disks to match the performance of the disk backup system, right?

In short, what differs and make special a disk backup system such as the Quantum DXi or any other, comparing to a standard SAN storage system, and how could achieve such fast throughputs?

Thank you,
Raphael

[chrisdearden]

The maximum speeds quoted by those systems would generally be the result of processing multiple streams of data. If you wanted to push your backup storage to its best performance you woudl nee to configure multiple backups over multiple proxy servers. Even then , try to avoid any reverse incrementals or synthetic full generation - its often quicker to take a new active full. While those systems can ingest data pretty quickly , they are not as fast to retrieve that data.

As for manufacturer quoted speeds , its a little bit like the fuel consumption figures quoted for a car - fine as an indication , but it can be difficult to repeat those figures in a real world scenario.

[veremin]

As to the bottleneck analysis, source bottleneck means that data cannot be provided by production storage to ESXi host any faster.

Furthermore, you might want to install Veaam ONE which is likely to give you real time stats regarding underlying storage appliance and so on. Even the free edition does provide several charts/dashboards that might be helpful for you.

Additionally, in order to increase performance it stands to reason to deploy additional physical proxy server and use it in Direct SAN mode.

Hope this helps.
Thanks.

[raphaelmsx]

Thank you for your answers.

Another quesiton regarding deploying additional physical proxy servers, if I do this, do I need to configure additional backup jobs on parallel in order to increase the performance, or just maintain the same single job running?

[foggy]

To benefit from having several proxy servers, you have to create several jobs so that backup tasks could be assigned to different proxies and performed concurrently. Note, that VMs in a job are processed sequentially.

[raphaelmsx]

As to the bottleneck analysis, source bottleneck means that data cannot be provided by production storage to ESXi host any faster.

Furthermore, you might want to install Veaam ONE which is likely to give you real time stats regarding underlying storage appliance and so on. Even the free edition does provide several charts/dashboards that might be helpful for you.

Additionally, in order to increase performance it stands to reason to deploy additional physical proxy server and use it in Direct SAN mode.

Hope this helps.
Thanks.

Hi,

you mean that 99% source bottleneck isn't entirely because of my production storage and proxy on a VM, if I deploy additional physical proxy servers in direct san mode, my production storage will be able to provide the data faster?

Thanks.

[veremin]

It’s just recommended in case of underlying FC storage to have physical proxy being specified in Direct SAN mode. In this mode VB&R retrieves VM data directly from the storage.

Thanks.