SD-Wan and Windows File Sharing/SMB not compatible?

[from_afar]

We installed SD-Wan a few months ago in a very simple single vlan spoke and hub setup--just the 2 locations, spoke and hub involved. Both locations has 1000/1000Mbps connections (verified both are getting expected speeds). When we try to copy a file from a Windows 2022 server on the hub to a windows 10 client (all fully patched) on the spoke, the file copy speeds (as well as iperf) peak at about 3Mbps. This makes it very hard for the users to get work done as the files their software uses are usually fairly large so everything takes a long time to open/save and files are getting corrupted regularly.

All AMP/IDP services are disabled on the spoke; tunnel all is on as we need all traffic to exit the HUB. We did all of the standard and even standard troubleshooting (disablebandwidththrottling registry edit, SMB tuning on the server, checking MTU, etc.) all to no avail. 

I went to the spoke location and scheduled a call with support (we get the service through ATT) who got a Meraki tech on the line. We spent 2 hours looking at things and testing with iPerf, testing file copy speeds between 2 machines on the spoke (worked fine) and at the end of it all, the final "answer" from the Meraki tech was that this is expected behavior with SD-Wan and SMB. If this is the case, how do people solve SD-Wan locations using fileshares? It seems like this should be a pretty standard and common setup, but is there really no solution to using SMB over SD-Wan (I confirmed we are using SMB3.1.1; packet captures show SMB2, SMB1 is definitely disabled). 

I was looking at maybe getting a better connection at the spoke (it is VZN Fios, so Fiber but shared) like DIA, but it's much more expensive and if SD-Wan is incompatible with SMB it wouldn't fix the issue anyway. 

[ww]

iperf shows the same low speed as smb?

or after help from support you get full/higher speed with iperf now?

[from_afar]

It isn’t always consistent and I’m not an iperf expert. I tried different window sizes and parallel streams. With the default settings for server and client, it does average around the same, though. That said, http downloads over the connection were a bit better (~10-15Mbps). 
iperf results have been consistent throughout—no faster after support call. 

Happy to run or share specific tests if you want to see them. 

[ww]

Are you using iperf3?

Should be very basic commands

Pc1 

Iperf3 -s

Pc2

iperf3 -c <ip pc1>  

Iperf3 -c <ip pc1>  -P 10

Iperf3 -c <ip pc1>  -R

[from_afar]

Thanks, yes, iperf3. Yes, those are the commands I ran (among others). The default settings (using just -c and -s) I get about the same rate--~3Mbps.

[ww]

Thats not much, dont think its just related to smb then 

Did you try ping a internet address for checking mtu on both sites?

Ping 1.1.1.1 -l 1472  -f

[cmr]

We have SD-WAN and use SMB (but v2) and don't have problems.  Our HQ has a 500/500 uncontended link and the DataCentre has a 1000/1000 uncontended link on the same carrier, however these are L2 MPLS (VPLS) and not DIA.

I regularly get 450Mb+ file transfers between the sites.  MXs are MX250 at DC and MX100 at HQ, all running 18.107.10, though they have been fine from 15.x upwards.

Are you testing at times of low activity and if so what is the utilization on the MXs at this time?

What MX versions are you running?

What models are at each end (I think you mentioned this in another thread?)

If you do an internet speedtest from each site, what do you get?

Are they on the same ISP (as I believe you have DIA and not private WAN?

How have you verified this: (verified both are getting expected speeds)?

[from_afar]

Thanks for the reply.

Are those windows file transfers? You said SMBv2 so I'm assuming yes, but just want to be extra clear.

Testing at all times and it never changes. This is a typical graph on our MX-95 (Hub):

I think MPLS line is the real differentiator here; we don't even have DIA at the spoke (but was looking into it). I don't understand why it would be different for MPLS if you are still using SD-Wan, you would think that whatever was causing my issue would happen on a MPLS link if SMB was really to blame.

I'm looking into Private Line now. I hate to spend the money, but this has been going on for a few months now and I have not been able to get it solved. 

[cmr]

I've got a spoke with an 80/20 line (VDSL) that has an MX68 on the end of it.  I'll run some tests to see how that performs, but yes, it was Windows (10 to Server 2016 or 2012 over the years).

[cmr]

I ran the Windows file copy tests today and got 72/17.5 Mb/s on the 80/20 line.  Both ends are running 18.107.10 and are MX68 at remote site and MX250 in DC, so there definitely isn't an SMBv2 issue with Meraki AutoVPN for me!

Could you please confirm these:

What MX versions are you running?

If you do an internet speedtest from each site, what do you get?

Is the hub also Verizon?

[from_afar]

Thanks for the reply.

I'm running 18.208 at both sites. The devices are set to auto-update on weekends, and this has been happening since I got the service around the new year (it took a while for the complaints to filter up and I was remoted in to a spoke machine and tried to copy a file and saw the awful speeds) at least the last couple of versions. It is scheduled to update to 18.211 next Sunday it looks like. 

I ran a speed test at the spoke directly on the Verizon equipment and it gave ~858/847. I do not have Verizon at the Hub--I have dedicated 1Gb fiber. It gives 836/869 directly connected to the MX-98. 

[BlakeRichardson]

I see you have an MX95 as a hub but what model are you using for the spoke and what is it's utilisation.

Also have you tried doing a copy from the server locally and confirm that it's not the servers NIC that isn't the issue.

[from_afar]

Thanks for the reply.

I have done local copies to test and get on average 100+MB/sec file copy speeds on the LAN between the server and same model laptops as are in use at the spoke. 

The spoke is an MX-68, and the utilization is very low (only 7-8 users) despite tunnel all mode. 

[PhilipDAth]

Have you got anything configured under Security @ SD-WAN/SD-WA & Traffic Shaping?

[from_afar]

Thanks for the reply.

No, everything that can be has been turned off including AMP, IDP, and traffic shaping. I did try adding Windows File sharing, setting the priority to high, but left DSCP tag alone but it didn't make a difference. 

[Dipen]

We do have file shares across our Network and we use it all the time plus we have HUB and SPOKE environment between 5 offices. MXs are far more capable of handling packets at least for us (since we have MX 105). Have you ran a test while transferring files what does the device utilization looks like since last time we had same problem and called tech support they insisted us disabling "AMP" and "IDP" under threat protection on HUB. In our case CPU usage was getting high and fans kept running all the time. That was my 2 cents hope this helps!

[from_afar]

Thanks, good to know. 

We are running MX at both location (95 at hub and 68 at spoke), with MS-120 switches going out to the clients. I'm getting the service through AT&T so I don't have full admin access unfortunately and I don't see anywhere where I can see utilization of CPU/RAM etc. the only charts I can see I have posted elsewhere in this thread. It doesn't look like it's using much on the traffic analytics pages, very little jitter and loss. 

[JGill]

I know this is an old post, but the CPU numbers are under the Organization -> Summary report (set the date / time to today if you want current view).  Or using the API call is an easy method to get 

 Sample Python script,  just set your org and api key.

import meraki
import os
import pprint

API_KEY = os.environ.get('MERAKI_API')
organization_id = os.environ.get('MERAKI_ORG')

dashboard = meraki.DashboardAPI(API_KEY, caller='TopApplianceUtil.py/v1.0')

response = dashboard.organizations.getOrganizationSummaryTopAppliancesByUtilization(organization_id, quantity= 500, timespan = 2000)

print ('Number of MX Appliances = ' , len(response))
for mx in response:

print(mx['utilization'], mx['model'], mx['name'])