Network perf issue (one way)

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

Rand__

Well-Known Member
Mar 6, 2014
6,634
1,767
113
While having fun with my esx boxes I moved all my vms to a backup ESX host so I can change EVC mode on my cluster. From the cluster to the backup box everything worked as expected (slightly slow but ok).

Now after I done the maintenance I wanted to move the VMs back only to find that my transfer performance is abysmal (~ 1-2 MB/s), Initially I thoughts its a vmware (drive) issue but it turned out - no its not.


Host A --- |-------------MLX 6036 |--MLX CX4----- esx 1-3
|(RJ45) | SFP+ (4:1 split) |
|-- Cisco SG350xG - 10G ethernet --SG350x| |
| (SFP+) | 1G RJ45 |
Host B --- ----------------------------------------------------


So i found out that my Backup ESX Host B has very bad performance when connecting to an IP thats associated with a box on esx1-3 on a dvswitch with a CX4 backing.
Host B is connected via SFP+ to a Cisco SG350xg, whcih in tun is connected to a SG350XMP via Ethernet, which is in turn connected to a SX6036 via SFP+ (4:1 splitter) and then to the target ESX Box via the CX4.
O/C everything is in vlans, mtu 9000.

Now if I run the same test via 10G network and UDP I also get very bad speed to ESX1
Now if I run the same test via 1G network and TCP I get line speed to ESX1
Now if I run the same test via 1G network and UDP I get very bad speed to ESX1

If I run ESX1 to ESX2/3 on any (>1g) network I get 20G+
If I run HostB <-> HostA on 10G line I get 10G speeds

I recently replaced the SG350x with an XMP modell (same capabilities, just + PoE) but I restored the config to the switch, so its configured identical...
I am sure this has worked in the past, but something is wrong and I can't find it...

I check duplex and mtu settings, vlan access/trunk setup and everything I could imagine, but to no avail.

Any ideas?
 

Rand__

Well-Known Member
Mar 6, 2014
6,634
1,767
113
So figured this one out ... (or solved it, have not determined exact root cause yet):

Brought up a second interconnect to the MLX switch with the Cisco - thankfully I run a 4:1 cable so plenty to spare.
Hooked it up the second SFP+ port and speeds are up to 5G. Not perfect, but definetly better then 20Mbit. Had some weird issues with the LAG I tried setting up, so took way longer than expected...

I think its the cable, probably bent it too much, but o/c also could be the port. Will update if I find out.