WAN Performance

[Mizinia]

I have an MX100.

I have been seeing this issue since I increased our bandwidth to 500 Mb/s the speed drops out about every 25 seconds when transferring large files. I usually notice this during cloud backups, but have seen it with other large file transfers also. I am looking for suggestions to try and resolve the issue since it drastically increases the time needed for the transfer and can cause the transfer to fail going to FTP sites.

[RaphaelL]

Hi , 

You would probably need to take a packet capture and see how TCP is handling / reacting.

Look for TCP zero and/or TCP window full. 

Do you see any packetloss on that link ?

Have you configured traffic shapping with the configured WAN speed ?

[Mizinia]

I do not see any TCP Zero or Window full messages. I do see a lot of "Ignored Unknown Record" messages in Wireshark.

I am seeing some packet loss but not on the same frequency as the transfer drop.

I do have traffic shaping configured but in this case the source has been whitelisted so it can use the full bandwidth.

[DarrenOC]

Hi @Mizinia  - what does the CPU look like on the MX?  How many clients do you have connected?

[Mizinia]

CPU doesn't look like it ever gets to 100% but it does get high (looks like >85%) during large transfers. the system shows about 450 clients. 

[RaphaelL]

In that case , that might be related to CPU usage. CPU must be spiking above 85% in that 1 minute window. The stats are tracked every minute if I'm not mistaken.

[Mizinia]

I had narrowed it down to needing a newer/larger appliance that was rated for higher throughput. I don't like the throw more money at it approach so I thought I would throw the question out to see if I was missing some configuration that would clear up the issue.

[RaphaelL]

100% with you there ! 

Do you have a way to test the transfer when the appliance is less utilized ? Let's say during the week-end or the night ?

[Mizinia]

i have tried with almost no users on. i see the same issue. 

[PhilipDAth]

99% likely you are seeing normal TCP windowing.

I wouldn't do anything about this ... you could try enabling TCP timestamps to see if it flattens it out a little bit.

netsh int tcp set global timestamps=enable

 Ideally you would do this on the machines on both ends.

[Mizinia]

I added the timestamps. we will see if that helps.