File server performance issues in ring topology layout

[Dusan Sladojevic]

Hello,

I would like to ask one question regarding performance issues of File server in a ring topology layout.
File server is connected to the first switch in a line.
Currently from the last two switches in a ring topology, when I download files from the File server I get the max speed of 20-30MBps.
From the other switches I get the speed of 60MBps.

We have a ring topology with 1st switch from above being WS-C3650-24PS,
and all the others being WS-C2960X-24PS-L.
All the switches have the same configuration.
Trunk ports that are connecting switches are configured like this:

 description xxx
 switchport mode trunk
 switchport nonegotiate
 load-interval 30
 spanning-tree guard loop

Server is connected to a port with configuration:

 description xxx 
 switchport access vlan 40
 switchport mode access
 spanning-tree portfast

Could the cause of these "slower speed" on last two switches be the ring topology? Because the File server is connected to the 1st switch in line?

[Mark Malone]

Hi

The more devices you pass through the more latency is attached each time which would slow it down but that seems like a significant jump 50% , did you investigate the 3rd switch to see if has any issues as that's the first switch to see the issue it may have some problems cpu , low on memory etc

Are the ping responses to the server from sw2 and sw3 the same

If not you could add a couple of port-channels in instead of single links to try and increase throughput  in case theres a heavier load in sw3 that's causing resources to be utilised more and cutting down on your filer speed

[Dusan Sladojevic]

I only have the results of ping response when server is not heavily used, and the ping response is the same - ranging from 1-11ms on all switches in this network (including the one on which File server is connected to).

Regarding the cpu and mem, as this models are the same - there are no difference on cpu or memory utilization..

Would in your opinion port-channel make any changes, nevertheless ping response is the same?

[Mark Malone]

If its possible it would be better in star topology break the daisy chain its never good to string devices off one another the last one will always suffer and never get what the first gets in reality ,I try an avoid it as much as possible , adding an extra link wont hurt either in etherchannel as its redundancy and bandwidth added , if you can spare the ports.

[Dusan Sladojevic]

Thanks for your reply Mark!
I will change the topology to star to see if this will resolve the issue.

[Joseph W. Doherty]

Which switch is the STP root?

[Dusan Sladojevic]

First switch in the line (C3650) is the STP root in this network.

[Joseph W. Doherty]

Ok, so the first two 2960 switches are one branch, and the bottom two 2960 switches another branch.  In other words, your hops are 3560 => 2960 (top on diagram) => 2960 (next under it); and 3560 => 2960 (bottom on diagram) => 2960 (next above bottom)?

[Dusan Sladojevic]

My network layout that I have presented is just like you said.
I forgot to mention that we have one more C3650 switch in other rack room that is connected over optical cables. You can see my updated layout.
Does this change anything?

[Joseph W. Doherty]

Does this change anything?

A little.  I've been trying to ascertain the actual path of traffic from your server to the various switches.  First reading of your OP, one might assume from your description the bottom two switchs' traffic has to transverse all the other switches above them.  But that might not be the case.  Could you verify what you STP topology is.  I.e how your traffic flows between switches.

Before you mentioned this other 3650, traffic might only be hopping through two switches.

[Dusan Sladojevic]

Hi Joseph,

I found that on switch which has this speed issue, has the port to the upper switch in block state:
Gi1/0/23 Altn BLK 4 128.23 P2p

So, yes all the traffic from this switch to the file server goes through the last switch in line (also affected by the speed issue), then over optical cable to the other rack room where is 3650 switch (not affected by the speed issue). and again to the main rack room over optical cable to the 3650 which is RSTP root, and on which File server is located.
Hmmm. still cant understand why do both switches experiance such a speed slowness when they traverse optical cable which is about 100m in lenght.
Do you have any more clue where to look at?

Thanks!

[pwwiddicombe]

60 Meg (bytes) is around 500 Meg, so probably what would be expected from a Windows server.

Are you sure all of your inter-switch links have consistent speed/duplex settings (all autonegotiate, or all set as 1000 Full)?  Mismatches can have drastic effects on speed.

Are you doing the testing during periods of "normal" or dedicated periods, or while there is heavy transfers other than your tests going on?  It might be you have other congestion or busy users at switch 3; and the uplinks 1-2 and 2-3 are heavily used, skewing results.

If possible, you might consider more of a star rather than a ring topology; particularly when the key resources are on one switch.  Links directly from switch-1 to the others can reduce effects of busy periods somewhat. 

However, don't forget redundancy that you might have had with your ring setup.  Use Etherchannel to have multiple links to the centrall switch, but make sure they have "diverse paths".  You will lose ports on the central switch, though, implementing this (as it will have multiple links to each (or critical) switches).

[Dusan Sladojevic]

Yes, you are right it is a file server running on MS 2012R2.
All switches are inter-connected over switch access ports with the same configuration I have already written. All switches are connected over CAT6 LAN cables.
The results I have presented, are when there is little or no traffic to this file server.
In the time when this File server is heavilly used, users on last two switches can experience speed drop to 750KBps.. :(
But in the same time speed drops on first three switches to 10-15MBps..
Guess this is the problem because File server is being connected to switch #1 only with one LAN cable. We will connect all 4 LAN cables from this IBM server to the network with Etherchannel/Link Aggregation. But this will not resolve the issue with speed difference from the last two switches.

Star topology was also my opinion, but just wanted someone else to confirm me if this could be the cause. We have enough unused ports on 3650 switch to support this star topology.

Do others have the same opinion that ring topology could be the cause for this performance drop?

[pwwiddicombe]

Have you verified the duplex does actually match all the way from switch-1 to the end?  "sh int counters errors" often brings to light problems you weren't aware of - framing errors, congestion, duplex mismatches, etc.  Look for the ports with higher error counts, and try to make sense of those.   Clear counters, and observe again in an hour or more.

Also, don't forget the server link(s).  Mismatched duplex or port-channels to servers tend to affect them greatly when many simultaneous transfers are in progress.

[Dusan Sladojevic]

Hi,

I have checked all the ports, and yes every trunk port in network is duplex.
sh int counters errors, didnt report any errors, only some outdiscards.
But i wonder, do those counters clear themselves automatically by time pattern..
I will investigate also this stuff when next traffic slowness to this file server occurs.
Thanks for the clue!