Replication or quickmigration performanc

[ivordillen]

Hello,

I'm setting up a veeam B&R environment with for the moment 1 physical server wich sees all the datastores (source and destination).

Because I installed 2 vm's as proxy (one at the source side - on the host that holds the source vm and 1vm as proxy on the destiantion side.)

I start a replication job and both are seen as hotadd but they don't become faster as 11MB/s while the vm's are both on a 6GB nic (vcflex10).

When I start the job with one physical proxy who sees all the storage I get source proxy as [san] and target proxy as [nbd].

kind regards,

Ivor

[foggy]

Ivor, transport mode selection is absolutely expected in both cases. Virtual proxies are capable of hotadd mode, which is more preferable than NBD and can be used to write data on target. When you select physical proxy with direct access to the source storage, Direct SAN mode is used to retrieve data from source, and as this mode cannot be used to write data on target, NBD mode is used for that.

Is speed your major concern here? Then please tell us the bottleneck statistics from your jobs report. Also, have you selected the proxy at the source side to store replica metadata, as explained at the corresponding step of the job wizard?

Thanks.

[ivordillen]

I expected a little bit more. In the setup where the one server did all the work I got a 16 MB/s average. The bottleneck is at the source side99% what is to expect because the nbd mode at the source side.
Because I have a one server setup and the repository is one of my backup repositories I don't think this is wrong?

I'm testing another setup and will show the numbers here.

You know I always want to be sure the setup is OK and I have tuned it for the best because when it's installed and in full production I don't have much time for such things and it's importing to have the best performance possible to keep the stress on the systems as low as possible.

thanks for the help.

[Vitaliy S.]

Hi Ivor,

Do you have 16 MB/s during full or incremental job passes? What is the make and model of your source storage?

Thanks!

[ivordillen]

My source storage is HP EVA8400 (216x600GB 15k) so shouldn't be the problem.
The bottleneck is always the target side wich is als EVA8400 with less spindels but should be fine also because I see nog stress on my controllers.

When I do backups from the one EVA to the other I always see the source as bottlenck. That's why I'm wondering if I'm doing something wrond.

The 16MB/s is during a full job.

kind regards,

Ivor

[foggy]

When I do backups from the one EVA to the other I always see the source as bottlenck. That's why I'm wondering if I'm doing something wrond.
The 16MB/s is during a full job.

Have you configured your source storage and proxy to use Direct SAN access mode to read VMs data?

[ivordillen]

Yes but I keeps getting nbd and the error that there are no available proxies running on the esx management interface.
My vm's are not in the same vlan but they can ping the ip adres of the management network.

Is the virtual appliance a possibility?
Could there be a advantage when I make the nics of the management network card faster? 2GB in stead of 500MB?

kind regards,

Ivor

[dellock6]

Ivor, veeam proxies can see all the datastores where the VMs are running from? That is, they are running on ESXi servers with those datastores correctly mapped?

Luca.

[ivordillen]

It all checks out fine so both proxies have san acces.

At the source side I see [san] and at the target side I see [hotadd] but the transfer speed is 6MB/s.

I see that the bottleneck is the target now but I really wonder what number I should expect in this case. (the replication is between 2 datastores on the same eva - I will check this with a datastore on my backup eva to see if this is beter)

What can I do to improve speed? What is a realistic number I should expect?

kind regards,

Ivor

[dellock6]

First we need to find out what is the maximum speed.
If you try to run IOmeter on the destination EVA, what numbers do you get? Try to use 4k chunks, 100% random, 100% write and see what comes out. I'm almost sure EVA can handle much more than those numbers, but just to be sure.
Also, there is maybe other traffic on disk activity on the destination EVA during backup operations?

Luca

[ivordillen]

It will have to wait for a week there I am going to Barcelona. I'll make some noise at the booth .

Ivor

[Gostev]

Better make some noise at our party on Tuesday night

[ivordillen]

I'll come and say something.

kind regards,

Ivor