Snapmirror Performance. (Single stream)

[dschroed33]

Hello,

I notice a conspicuousness that a Snapmirror connection does not come over 300 MBs. Both systems FAS8200 & Fas8080 are connected with 40 GBits via a LACP. Bottlenecks in the network are excluded and the hard disks are at 10-20% utilization.
Is there an option which controls this?

Snapmirror throttle is unlimited. Currently ONTAP 9.1P7

Many Thanks
Dennis

[AlexDawson]

Hi there,

Can you run this command on all nodes:

node run -node * options -option-name replication.throttle

and paste the output?

That said - in general it depends on how much other workload is present on both systems, what sort of disks are underneath - and at the end of the day it is a lower priority task.

For detailed troubleshooting, you should open a case with our support center - you can do that via the web on https://support.netapp.com

[GidonMarcus]

Hi.

what do you call a single stream? and what the latency between the two sites?

Note that if you using the 40GB card and/or lacp.. you better create multiple intercluster LIFs per node. otherwise the traffic will go via a single 10GB path.

Have you tried multiple snapmirror at once. or only single volume? if it's a single volume, there could be a serialization bottleneck, that could be picked up in perfstat and you may get better result on XDP then DP  type (Defualt in 9.3, so worth migrating to anyway).

Gidi

[kahuna]

 - on a single volume workload (such as snapmirror), you will hit a CPU bottleneck way before the 10G limit so don't focus too much on the network

 - what is the throughput you expect?

 - as already mentioned here, snapmirror operates under low priority compared to user workload. Once a certain CPU threshold is crossed, snapmirror will be throttled. The only way to change that is with a special flag

 - if you'd like to further investigate the issue, we will need a perfstat (as already mentioned) from the source and destination during the entire snapmirror transfer

[symstor]

Hi,

Although you have a 40Gb pipe, are they arrays within the same DC or split over two DC's?

Kind regards

Andrew Martin