Ask the Expert: Access LAN Switches (Cisco Catalyst 4500E, 3750-X, 3560-X, and 2960)

[ciscomoderator]

With Nikolay Karpyshev

Welcome to the Cisco Support Community Ask the Expert conversation. This is an opportunity to learn and ask questions about architecture and troubleshooting Access LAN Switches such as Cisco Catalyst 4500E, 3750-X, 3560-X, and 2960 with Cisco Expert Nikolay Karpyshev.

Nikolay Karpyshev is a Customer Support Engineer in the high touch technology support  team (HTTS)  at Cisco specialized in LAN Switching. Karpyshev supports the Cisco Switches Nexus 7000, Catalyst 6500, 3750, 3560, 4500, 2900, among others, and works as senior and escalation engineer. He was previously a part of Cisco Sales Associate program. He holds a specialist degree in Mathematics and Mechanics from Novosibirsk State University in Russia. Nikolay also holds these Cisco Certifications:  CCNP, CCSP, and CCDP.   

Remember to use the rating system to let Nikolay know if you have received an adequate response. 

Nikolay might not be able to answer each question due to the volume expected during this event. Remember that you can continue the conversation on the Network Infrastructure sub-community discussion forum shortly after the event. This event lasts through July 27, 2012. Visit this forum often to view responses to your questions and the questions of other community members.

[elguen.hasanov]

Hi Nikolay,

Thank you for your answer.

Have a nice day.

Elguen

[manumohan200]

Hi Nikolay,

I have a customer having a VDI server infrastructure. He is connecting his VDI servers to a Catalyst 6500 switch with WS-X6548 module.  With this setup, he is having some performance issues where the VDI application is functioning very slowly. This happens in any of WS-X6548 modules we have in our network.

We changed these servers to connect to our Catalyst 4500 switches in their Ws-X4248 module. After this there was no application issues.

Do you have any difference in terms of the buffer size or the througput capacity of these line cards. If yes, what are those values?  Please revert.

Thanks,

Manu

[Leo Laohoo]

I have a customer having a VDI server infrastructure. He is connecting his VDI servers to a Catalyst 6500 switch with WS-X6548 module. 

Servers connected to a 6548 line card????

The 6548 was NOT designed to be used by servers.  The 6548 was designed as a desktop line card (hence the PoE daughter card module as an option).  Servers are recommended to use the 6748 line cards.

The use of 6548 connected to servers will cause the line cards to drop packets due to buffer overflow.

*** Sorry Nikolay for hijacking this post. ***

[manumohan200]

Hi leolaohoo,

Thanks for the reply. I got your inputs already from a Cisco URL. I understand that the buffer size of the 6548 line cards is 1MB per 8 ports.

I want the same parameter for  4500 line cards., ie buffer size of each of the ports in WS-X4248 line cards. I am not getting this informatiom from any of the Cisco documents?

Thanks,

Manu

[nkarpysh]

Hi Manu,

I'm a bit confused. Afaik the 4248 card provide only 10/100 speed for ports. And the buffer of 1MB per 8 port you talk about are specific to 1 GB per port 6548 line cards only. 100Mb 6548 line cards have different buffer specifications. Did you have a server on 1GB port having a problem and then moved to 100MB port solving the issue?

Can you specify exact line cards you used for me to come with the correct answer for you.

Nik

[manumohan200]

Hi Nikolay,

This is the setup.

I have some set of servers. If we connect to 4248 module, there is no issue with the application.

If I connect to 6548 module, the application is slow.

From one technote related to WS-X6548 module, I understand that 6548 modules are not suitable to connect the servers, but only end computers.

I would like to know the technical reasons behind this in terms of some values (say some buffer sizes) so that I can convince my customer to go for some higher module for server connectivity.

Thanks,

Manu

[nkarpysh]

Hi Manu,

If you talk about 10/100 6548 LC like WS-X6548-RJ-45 then it has 1.2MB of buffer per port and is designed for Server Farms.

If you talk about 10/100/1000 card like WS-X6548-GE-TX then it has 1MB per 8 ports as you said above and is designed for Gig to desktop. Other thasn buffer restriction this card has 8:1 oversubscription to fabric and that is a major problem for servers.

http://www.cisco.com/en/US/partner/prod/collateral/switches/ps5718/ps708/product_data_sheet0900aecd8017376e.html

The reason why this card is recomended for desktop or IP Telephony is that those end stations rarely goes up to the line speed thus usually don't need to queue traffic and use the buffers. They also can share this speed easily with other ports as desktops and Ip Phones rarely start to talk all at same time at big speed.  Thus buffers are quite small here and oversubscription is present.

In case of servers -those can get multiple connections and easily reach the line speed - thus queueing will be needed and so Higher buffers are requirement and also non-blocking connection to fabric.

Thus all depends about the type of card you talk which I still did not get

Nik

[manumohan200]

Hi Nikolay,

Thanks for the response.

I understood the limitations on 6548-GE-TX module.

Can you tell me what is the buffer size per port for WS-4248-RJ-45 module and the subscription rates?  I am not able to see this values in any of the Data Sheets.

Thanks,

Manu

[nkarpysh]

                   Hi Manu,

Afaik 4500 use different buffering model. Supervisor provides a certain buffer space (depending on SUP version) for all line cards and ports. Thus that buffer space will be dynamicaly spread among all line cards/ports. So single port can get from few hundreds of bytes up to several MB if no other need it.

In regards to oversubscription - 4248 10/100 MB LC has 6 GB connection to SUP thus not oversubscribed at all.

Thus to your situation. It can be that you connected server to 6548-GE card first which has already some servers connected to the same port-group. Those servers might have eaten the 1GB badwidth already thus the new one had to compete even for fw megs.

When you moved to 4248 you provided dedicated 100MB link which solved a problem. So it may be related to over-subscription to backplane only and did not come to buffers yet.

That is just  guess and one of the posibilities based on the symptoms discibed.

HTH

Nik

[njackson40416]

Hi Nikolay,

I have another question. We have a computer that is going into error-disable due to a link flap.

The thing that is confusing us is the port security only trips on Monday evenings. This has been ongoing for over a month now.

Cables have been replaced, but it still confuses me as to why the error only happens every Monday.

We have automatic updates, could that be a cause? Or perhaps a misconfiguration of IPv6?

Any help would be greatly appreciated.

Thank you

V/r

Chai

[shillings]

Hi Nikolay,

What can you tell us about the future of the current 10/100 2960 series switches?

It's has a good lifespan already, so any plans you can share with us?

Many Thanks,

[nkarpysh]

Hello,

2960 is still in and developing. 2960-s is one of the new platforms which will be there on market for a while. Here you can find a list of Q&A for this platform:

http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps6406/qa_c67-577519.htmlhttp://www.cisco.com/en/US/partner/prod/collateral/switches/ps5718/ps6406/qa_c67-577519.html

http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps6406/prod_qas0900aecd80322c37.html

And the plans are to grow further. Let me know if you have any specific question and I will help you with it.

Nik

[nkarpysh]

Hi Chai,

The thing to start with  - please tell me what is the switch you have a problem on. Also here port is error-idsabled not by port-security but by the non-stable link factor to keep switch control protocols safe of it. The next thing to consider - what is connected to that port and what kind of traffic going through it at that time. Is it copper or optical port? Is there any patch pannel in between?

For the first tests to do I would recomend following:

- Try different speed/duplex settings on that port:

--- speed/duplex hardcoded to 100 ro 1000 (if supported)

--- spped/duplex configured to auto

--- speed negotiation disabled

Nik

[Leo Laohoo]

Nikolay Karpyshev wrote:

Hi Chai,
The thing to start with  - please tell me what is the switch you have a problem on. Also here port is error-idsabled not by port-security but by the non-stable link factor to keep switch control protocols safe of it. The next thing to consider - what is connected to that port and what kind of traffic going through it at that time. Is it copper or optical port? Is there any patch pannel in between?
For the first tests to do I would recomend following:
- Try different speed/duplex settings on that port:
--- speed/duplex hardcoded to 100 ro 1000 (if supported)
--- spped/duplex configured to auto
--- speed negotiation disabled
Nik

How about doing a TDR to check if Chai has a Layer 1 issue?

[nkarpysh]

Thats is a good idea Leo.

However we still first need to find out what is the switch used and if it has TDR in built. If we consider external TDR then that wont be of much use I guess as the cable was replaced already. Anyway that will be still good to have the results of such tests during the problem and normal work.

So Chai - you can talk to your site team to include these tests on their radar.

Nik