Slow File transfer in one direction and Fast transfer in opposite direction.

[earl.mceachron]

A Windows 10 Enterprise desktop connected to a C4506-E (1G) takes about 2 hours to transfer a 2.2G file to a VM (VMware)  running Windows Server 2012R2 whose host is connected to a Cisco 9372PX FEX port (10G). Reversing direction, where the same file when transferred back to the desktop only takes about 20 seconds.

Desktop to any other server including VMs, the process takes less than 2 minutes max. This process is only slow from a physical desktop connected to the C4506E transferring to this particular VM. By the way both machines are on the same VLAN.

I upgraded the C4506E SUP8L-E from 03.09.01.E to 03.11.04.E; and the Nexus9000 C9372PX from nxos.7.0.3.I7.7 to nxos.9.3.7

Has anyone seen this phenomenon before? Any ideas anyone?

[balaji.bandi]

Do you see any MTU Mismatch ?

Do you have any QOS ?

what about transferring between the devices connected to same switch ?

how about same VM to another VM transfer ?

check the interface drops any on the switch side connected devices?

[earl.mceachron]

Do you see any MTU Mismatch ? no all the ports are set at 1500

Do you have any QOS ? No QOS configured

what about transferring between the devices connected to same switch ?  Normal speed much faster than to the specific VM Server

how about same VM to another VM transfer ? Going to the specific VM Server is slow. the return is fast.

check the interface drops any on the switch side connected devices? No drops all interfaces showing 0 drops

[Sergiu.Daniluk]

You will have to narrow down the problem first o a switch (N9K or C4500). To do that, check what is the transfer rate for some traffic which is passing ONLY through N9K. If you do not have something, simply connect your laptop to the N9K, and test the transfer rate with the VM. If you do not see any issues, then the problem is C4500. If the rates are slow, then you need to focus further on N9K.

The problems with slow transfer rates are:

+ packet drops/discards, because of buffer exhaustion

+ CPU processed traffic, instead of ASIC switching (N9K specific)

+ CRC errors which leads to retransmission's

+ low MTU on the path

Stay safe,

Sergiu

[earl.mceachron]

 Not as straight forward as you may think. I did all the suggested checks. Transfer in one direction is fast. The reverse is slow. That's the kicker.

I finally decided to check the datastore that the specific VM resides because the only slow transfer takes place when transferring to that VM. Transferring from that VM is fast. There is a datastore error. I am having my storage engineer look into that. I will update this post when that error is fixed.

Thanks!