Physical and transport layer issues

[sdrcpdpune]

During network diagnosis I found that there are no errors in "ifstat -a" output but there are some issues with in the physical and transport layers in "netdiag -v" in our both filers. Below are the output of that:

---------------------------------------------------------------------------------------

Performing physical layer diagnostics.....

The interface e4a seems to be receiving an excessive number of packets for protocols not supported this system.

If this problem persists, please contact your network administrator.

The interface e4b seems to be receiving an excessive number of packets for protocols not supported this system.

If this problem persists, please contact your network administrator.

Performing network layer diagnostics.....OK

Performing transport layer diagnostics.....

The TCP receive window advertised by NFS client xxxxxx is 6912.

This is less than the recommended value of 8760 bytes. You should increase the TCP receive buffer size for NFS on the client.

Average size of CIFS TCP packets received from host: xxxxxx is 223.

This is less than the MTU (1500 bytes) of the interface involved in the data transfer.

The maximum segment size being used by TCP for this host is: 1460. Low average size of packets received by this system might be because of a misconfigured client system, or a poorly written

client application.

There are more than 100 messages like this from both filers. (I have cluster of two 3140 nodes)

---------------------------------------------------------------------------------------

I have few queries on above issues:

1. How to check the physical layer problems which is marked by red color?
   
2. The transport layers shows the wrong buffer size from both unix and windows and that count of unix and windows is 100:10. Does the fixing of right buffer size in windows client improve the performance of that client or whole filer?
   

If there is any document on that to read about, it will be helpful.

[sdrcpdpune]

What??

Is this problem not faced by anyone here?

OR if possible can anyone take this to Netapp Engineering team, I tried from my end  by opening a case, but the folk who took it is too slow and trying to give solutions which are not relevent to this problem.

[chriszurich]

I'd start by confirming the MTU size set for each IP interface on the filer.  Launch FilerView and navigate to the Network section and select Report.  In the right hand field you'll see each interface with the second column displaying the configured MTU size.  All interfaces should be set to 1500 unless you are using Jumbo frames in which case the interface will be set to use an MTU size of 9000.  There should be no deviation from these two values. Ignore the MTU size (8160) set for the lo.

If any of the interfaces are not set to 1500, update the interfaces by selecting manage from the left column and then select each interface and set it's corret MTU size, 1500 in most cases.

I would also confirm that the clients are connecting to the filer on the right interface IP.  Go to each client listed within your error and confirm what interface IP it's CIFs share is connecting on.  The same troubleshooting step should be performed for the NFS connection errors in question.

I also assume that you've confirmed that CIFs & NFS are licensed on the filer and that your license has not expired?

[aborzenkov]

The red color messages have nothing to do with NetApp. NetApp supports TCP/IP suite; may be your network includes IPX, AppleTalk, NetBUI (hopefully, not ) or anything else and this traffic leaks to NetApp. If it of concern, you have to sniff traffic on ports connected to NetApp and analyze it. The first step would be to sniff traffic on any port in the VLAN and check for excessive non-IP broadcasts.

Regarding second warning - it has been discussed in length recently here (do not have reference handy). Basically, different NFS/CIFS operations have different length, most of them being far smaller than 1500 bytes. I am not sure anything can be done here.

[sdrcpdpune]

Hi,

Thanks dudes,  I checked all interfaces MTU size, all are set to 1500 except lo, which is ok.

Take eg of this error:

The TCP receive window advertised by NFS client pnli6036.net.plm is 6912. This is less than the recommended value of 8760 bytes.
You should increase the TCP receive buffer size for NFS on the client.

My query is :

Do these kind of messages impacting the performace of the filers?

Do I need to set the required TCP window size on each and every unix and windows machine which are reported by Filers?

If uers are facing slow performace issue of windows machines (CIFS sessions), then fixing only windows machines TCP window size improves performance, leaving the unix machine's TCP window as it is?

regards