cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
4520
Views
0
Helpful
8
Replies

Qos on Switchport

neteng2323
Level 1
Level 1

I currently have the below configuration on a switch and switchport. M question...

Question1:  If this port is connected to a server (no phone) should I remove the Qos settings on just this port?

Question2: What happens with data only traffic with the current 40 20 20 20 share, and priority-queue out command?

!

mls qos map cos-dscp 0 8 16 24 32 46 48 56

mls qos

!

switchport access vlan 30
switchport mode access
switchport voice vlan 60
srr-queue bandwidth share 40 20 20 20
priority-queue out
mls qos trust device cisco-phone
spanning-tree portfast
spanning-tree bpduguard enable

2 Accepted Solutions

Accepted Solutions

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages wha2tsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Ok, that helps, but still don't know your traffic or the rest of the device configuration.

Highly unlikely, for a non real-time device, PQ being present, or not, matters little to the host.

I believe the 2960s, like 3560s and 3750s, is very sensitive to their buffers configuration.  (Not positive about the 2960-X, but other 2960s are noted as only having 2 MB buffer RAM, shared by all the switch's ports.)

Also, I believe basic default buffer settings, when QoS is enabled, often provides less buffers then when QoS is globally disabled.  (NB: a "tuned" configuration, can optimize performance over both default and disabled QoS.)

I would suspect, due to lack of buffering, packets are being dropped.  This could be more true, because other devices are running at gig, but this server is running at 100 Mbps(?).  If so, its easy for a gig to overrun FE.

View solution in original post

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages wha2tsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

When you enable QoS, globally, by default, a fixed amount of buffer space (usually 50%) is reserved (and split again) for each of the four interface hardware queues.  The remaining space is placed into a common pool, that interfaces can "borrow" when they run out of their dedicated/reserved buffer space.

When it comes to buffer management, I believe the 2960 has one of two queue-sets that you can assign a port to (by default, ports use queue-set 1).

Buffers, as allocated to hardware interface queues, are also based on percentages, of percentages.

On 3750s, I've found by decreasing interface buffer reservations, and/or by increasing the maximum allocation a hardware queue can draw, I often dramatically decrease drops on transiently congested ports.

e.g.:

mls qos queue-set output 1 threshold 1 1600 1600 25 1600

mls qos queue-set output 1 threshold 2 1600 1600 25 1600

mls qos queue-set output 1 threshold 3 1600 1600 25 1600

mls qos queue-set output 1 threshold 4 1600 1600 25 1600

View solution in original post

8 Replies 8

Reza Sharifi
Hall of Fame
Hall of Fame

If the port is used only for a host (PC, server, printer, etc) than there is no need for the QOS setting.  Its just extra config you don't need.

here is good link on queuing regarding your second question:

http://www.cisco.com/c/en/us/support/docs/switches/catalyst-3750-series-switches/91862-cat3750-qos-config.html#topic3

HTH

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages wha2tsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

If the port is used only for a host (PC, server, printer, etc) than there is no need for the QOS setting.  Its just extra config you don't need.

Cringe.  ;)

QoS covers a lot, often more than one might think of at first glance.

For example, even for a host like a printer, assuming all output is just "print" data (which, BTW, it might not be, as many modern network printers might support, across the network, working with the printer's "control plane"), QoS might be used to address "goodput" (https://en.wikipedia.org/wiki/Goodput).

To the OP's two questions . . .

#1 Insufficient information to say

#2 Also insufficient information to say

The port needs a very basic QoS config.  The Linux server that is connected to the switch is old, and has only a 10/100 NIC.  It does some file transfers from time to time and runs some software that controls tools.  

Here is a little backstory on this...

We recently upgraded the network closet where this server connects.  Before, there was no QoS on the old connection\switch, and after the upgrade the admin that oversees this server complained that file transfers to it had slowed down alot.  Transfers out (ie to another server) were fine. I found the above config, but it was missing priority queue out and I noticed a lot of output queue drops.  I haven't had a chance to have them test it again after adding that commmand, so I'm curious if it will fix the issue or if I need to look deeper into the qos config?

What's the platform and IOS version?

2960-X Stack, IOS 15.2(2)E3

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages wha2tsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Ok, that helps, but still don't know your traffic or the rest of the device configuration.

Highly unlikely, for a non real-time device, PQ being present, or not, matters little to the host.

I believe the 2960s, like 3560s and 3750s, is very sensitive to their buffers configuration.  (Not positive about the 2960-X, but other 2960s are noted as only having 2 MB buffer RAM, shared by all the switch's ports.)

Also, I believe basic default buffer settings, when QoS is enabled, often provides less buffers then when QoS is globally disabled.  (NB: a "tuned" configuration, can optimize performance over both default and disabled QoS.)

I would suspect, due to lack of buffering, packets are being dropped.  This could be more true, because other devices are running at gig, but this server is running at 100 Mbps(?).  If so, its easy for a gig to overrun FE.

That makes sense and I was also aware that the 2960 buffer is limited.  I want to make sure I'm understanding you correctly, in this case I have Qos on the port, so the buffer size is smaller and could be contributing to the problem.  Would removing the settings on the port itself give me the default buffer size?  The server is indeed running at 100 Mbps and the complaint is only file copies TO the server from servers that are on different networks and on 1 and 10 Gbps interfaces.

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages wha2tsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

When you enable QoS, globally, by default, a fixed amount of buffer space (usually 50%) is reserved (and split again) for each of the four interface hardware queues.  The remaining space is placed into a common pool, that interfaces can "borrow" when they run out of their dedicated/reserved buffer space.

When it comes to buffer management, I believe the 2960 has one of two queue-sets that you can assign a port to (by default, ports use queue-set 1).

Buffers, as allocated to hardware interface queues, are also based on percentages, of percentages.

On 3750s, I've found by decreasing interface buffer reservations, and/or by increasing the maximum allocation a hardware queue can draw, I often dramatically decrease drops on transiently congested ports.

e.g.:

mls qos queue-set output 1 threshold 1 1600 1600 25 1600

mls qos queue-set output 1 threshold 2 1600 1600 25 1600

mls qos queue-set output 1 threshold 3 1600 1600 25 1600

mls qos queue-set output 1 threshold 4 1600 1600 25 1600