Slow performance after 30-60 minutes runtime

[SAPNWADM]

Hello,

we have a cluster with 4 ESXI 5.5 Servers and several SAS attached HP P2000 G3. Our repository is a HP Windows 2012 R2 Server with a 10 Gb 530 T2 Card. The backup Network ia a 10 Gb network with a Netgears XS 708E Switch. The repository is a HP D2600 with 12 SAS drive (7.200 rpm) connected via Smart Array P421 Controller)

The Fullcusterjob contain 158 VMs with 36 Tb data. Each ESXI-Server has its own Veeamproxy VM connected to the 10 Gbit network.

When the job stars, we get a processing speed from 1-1,6 Gb/sec with a transfer rate form 400-700 mb/sec. On our repository server i see a network speed from 4-7 Gbit/sec (Windows Taskmanager) and the Harddisk writes with 400-500 mb/sec. Bottleneck is "Source".

After 25-60 minutes after starting the job, the Network perfromance decreases. The network speed goes up and down from 5 Gbit/sec to some Mbits for some seconds. Then going up to 5-6 GBits. The phase with high networkperformance will get shorter and shorter and then the bottleneck changes to network.

But if i test the throughput of the vproxy servers and the repository server with iperf i will get the same performance as i get before starting the job.

Thanks

[dellock6]

Hi,
are you using v8 Update 2 already or a previous version? At first sight, and looking at the size of the backup file (it's a single backup job handling all the VMs, correctly?) seems to me like you are hitting the repository cache too much and this has to flush data from time to time, especially after the job is running for some time. Out of curiosity also, how much RAM is installed in the windows repository server?