High percentage of retrasmit on proxy NIC

[maol]

Hello Guys,
I've just noticed a trange beawiour on my new veeam setup.
I've two physical proxy accessing production storage in direct SAN and both acting as a gateway for a DDBoost over fc repository.
Sometimes some session have the data ridden from proxy A and written by proxy B - this is expected.

both proxy are connected to a 20Gbit LACP PO and iperf is performing correctly.

when this situation happen i saw lots of discarded packet and retransmit at the NIC level:

Code: Select all

PS C:\Users\f56945a_adm> netstat -e
Interface Statistics

                           Received            Sent

Bytes                    1475080611      1108576590
Unicast packets          2652122594      3973294747
Non-unicast packets         1330555           26870
Discards                    5452026               0
Errors                            0               0
Unknown protocols                 0
PS C:\Users\f56945a_adm> netstat -s -p tcp

TCP Statistics for IPv4

  Active Opens                        = 2060
  Passive Opens                       = 951
  Failed Connection Attempts          = 39
  Reset Connections                   = 1119
  Current Connections                 = 107
  Segments Received                   = 264737393
  Segments Sent                       = 709599594
  Segments Retransmitted              = 83710061

but the backup is perfoming well, with about 600MBps average throughput.
I'm assuming that discarded packet might be caused by the fack that reding from the production storage is way faster that writing on the dedupe appliace,
some buffer eventually get full and the receiving proxy start discarting packet.

Can anyone confirm that this is an "expected behaviour" and we can ignore this?

Thanks
L

[PetrM]

Hello,

Looks like it is out of our control as the issue occurs on a lower layer of the OSI model. Maybe you could check the NIC driver version but I believe you can ignore it as long as our jobs do not report any errors and fit your backup window.

I think it's very specific to your infrastructure and I really doubt that this is the expected behavior. Maybe you can try to copy a huge file from proxy A to proxy B to see whether you have the same NIC statistics? Another idea is to run network traffic dumps (Wireshark/tcpdump) on both proxies and analyze them, maybe it will shed light on it.

Thanks!