FTP across the campus lan

[ohareka70]

Hello,

I am looking at the performance of the campus lan from 5 buildings going to the file servers in the Data Centre.  Consistently from all the buildings downloading a 2.2Gb file takes around anything from 5 - 9 minutes downloading and 4-5 minutes uploading the same file and to the same server.

From a routing/switching point of view is their anything i should be looking at that might be affecting the downloads. Is their a reason why uploading is faster that downloading.  Or is this more server related than cisco networking related?

any advice is welcome

regards,

Kevin

[Richard Burts · Accepted solution]

Kevin

Given what we know so far I would think it more likely that it is an issue on the server and less likely that it is a network issue. It might be interesting if you could do a transfer to and from the server from a device in the data centre, perhaps even in the same subnet, so that you get most of the network out of the way. It seems this would be a way to see server performance separated from network performance.

I can think of a couple of things that might have impact on the transfer from a networking perspective. To understand these it may be helpful to remember that in an FTP transfer there is a large flow of big packets (the data) in one direction and a smaller flow of small packets (the acknowledgements) in the other direction. If there were some network issue that had more impact when data flows from the server and had less impact when data flows to the server, then it might produce the effect that you are seeing. Here are a couple of things that might have more impact in one direction and less impact in the other:

- if there is a duplex mismatch (device on one side operating in full duplex and device on the other side operating in half duplex) then the full duplex device will happily transmit but the half duplex device will be seeing collisions (and late collisions) resulting in re-transmission. So you might check for some device in the path outbound from the server that is operating in half duplex mode.

- if some interface were enforcing a smaller MTU it might create a need to fragment traffic in one direction and impact traffic.

It might help to find an answer if you could do a packet capture (Wireshark or whatever packet capture tool you may have) on some of these file transfers. It could show you if there were fragments in one direction and not the other, or if there were lots of re-transmissions in one direction and not the other.

If you were seeing symptoms in one or some building but not in others I would be more likely to think it was a networking issue. Since the symptoms seem to be pretty uniform in all buildings I would think it less likely to be a networking issue.

HTH

Rick

[ohareka70]

Rick,

Data transfer between this server and another server on same subnet for 2.2Gb is 3-4 mins.  Across the lan/wan it is around 8-9.

I have checked the duplex settings on the stack and 3 servers were mac flapping so we brought them offline, teamed the network cards again and set them for etherchannel.  The mac flapping disappeared and there is definitely some improvement in the performance of the servers. 

We plan to upgrade the ios on the stack later to match the ios version on a stack in another building.  They have a 10Gb link between them and we think their ios versions should match.

We are thinking of create a vlan on every building (we have over 200 users across the campus lan) but not all the switches are vlanned.

we plan to replace the 2 file servers with new ones with 8Gb ram and 15k sas discs.  Currently they are running 2Gb ram and 10k scsi discs.

thanks

Kevin

Just one more question - i have netflow (manage engine) running across the lan/wan on the 2811 routers to see whats going on.  But i havent been able to get it running on a layer3 switch yet.  Do you know if this is possible?  i 'd like to get his running on the 3750 layer stack switch in the datacentre to see what bandwidth is being used on the servers by the users.

[glen.grant · Accepted solution]

Netflow is not supported on a 3750..

[Richard Burts]

Kevin

I am glad that you have found and fixed some issues and that performance has improved. I believe that upgrading the IOS to match is a good thing, but I am not sure if that would impact performance. I do not know enough about your environment to be able to say much about separate VLANs for each building, but in general I would support separate VLANs, especially in terms of being able to limit the impact of a problem that might occur within a building. I would guess that replacing the file servers would be the most effective way to improve performance.

Glen is correct that Cisco does not support NetFlow on 3750 switches. For some details about that you can use this link:

http://www.cisco.com/en/US/prod/collateral/iosswrel/ps6537/ps6555/ps6601/prod_white_paper0900aecd80406232.html

HTH

Rick