Solved: Re: RSTP or UPLINKFAST instead of standard STP

socrates_2000 · ‎09-23-2010

We have 3 sites, each sites having 28 servers , 10 3550 swittches and 2 3750 switches.

Each server has 2 NIC cards. There are 2 Vlans, each server belong to different VLAN. Each VLAN port of server is directly connected to 3550 switches in order to connect it to IP network. These 3550 switches are in turn connected to 2 redundant 3750 switches.

To route the traffic from a site to another, the 3750 switches are running at L3 using HSRP and EIGRP. For intra-site communication, our 3550 switches and customer 3750 are running at L2 using standard Spanning Tree 802.1D

In each site two 3750 switches are root bridges of VLAN1 and VLAN2

Because of std STP the convegence time is too high around 50 seconds , and its causing problems in the server and database synchonisation.

The possible solutions we have is:- RSTP and enabling uplink fast on 3550 switch.

We can do this only in one site at a time.

Please help me in the following:-

1) What are the risks and conseqences whilie impleneting RSTP and uplink fast.

2)Which of the 2 solutions is better or any better solutions, and how can we implement this, without causing any outage?

Jon Marshall · ‎09-23-2010

2) Make sure that, all the trunk ports of 3550 and 3750 switches are operating in Full duplex mode. Or do i have issue the command to force the port as point-to point link, ie "spanning-tree link-type point-to-point" ? Which is the safest option?

By making sure they are full duplex the switch should see them as P2P links.

Please confirm if this proceedure is correct? Usually how long will it take for the RSTP to recalculate parameters and attain stability assuming that there is no topology changes? Anyways am going to request the customer for 1 hour maintenance window.

Procedure looks fine to me. To be honest, assuming you don't have to reload switches, we are talking minutes for it reconverge to a stable network. I suggested an hour because you just never know what might happen and not necessarily with STP ie. you may reload a switch and it does not come up correctly etc. I am not trying to worry you but it's always best to plan for all eventualities.

To enable uplinkfast in all the 10 switches , i have to issue the command "spanning-tree uplinkfast" in global configuration mode right? (or i have to issue the command in interface level?). I have already enabled "portfast" in access ports fo 3550 switches that is connected to the servers. Does Enabling "uplinkfast" requires it to be changed or i leave it alone? I feel it shouldnt be changed, but still i want to make sure.

You should enable in global config mode. You should leave portfast configured ports as they are. You obviously don't want portfast configured on the switch uplinks.

To implement uplinkfast If i issue the command "spanning-tree uplinkfast" in all 10 switches one after another, how long will it take the network to be stable?

Again, if there are any STP calculations we are talking minutes at most for it to reconverge.

Most probably customer is going to agree only for UPLINKFAST.

If you are going to be getting an outage anyway, and outages are not that easy to come by, i would strongly recommend you advise the customer to go with RSTP and not just uplinkfast. RSTP is a great improvement on STP and not just because it incorporates uplinkfast. If all your switches support it there is no reason not to migrate to it.

BTW its good to know you were a Solaris admin, I am also from the same domain specializng on SUN hardware and Solaris.

I still miss working on Solaris sometimes but i suspect if i was put in front of a Solaris server now i wouldn't haven't a clue what to do

Jon

View solution in original post

Jon Marshall · ‎09-23-2010

socrates_2000 wrote:

We have 3 sites, each sites having 28 servers , 10 3550 swittches and 2 3750 switches.

Each server has 2 NIC cards. There are 2 Vlans, each server belong to different VLAN. Each VLAN port of server is directly connected to 3550 switches in order to connect it to IP network. These 3550 switches are in turn connected to 2 redundant 3750 switches.

To route the traffic from a site to another, the 3750 switches are running at L3 using HSRP and EIGRP. For intra-site communication, our 3550 switches and customer 3750 are running at L2 using standard Spanning Tree 802.1D

In each site two 3750 switches are root bridges of VLAN1 and VLAN2

Because of std STP the convegence time is too high around 50 seconds , and its causing problems in the server and database synchonisation.

The possible solutions we have is:- RSTP and enabling uplink fast on 3550 switch.

We can do this only in one site at a time.

Please help me in the following:-

1) What are the risks and conseqences whilie impleneting RSTP and uplink fast.

2)Which of the 2 solutions is better or any better solutions, and how can we implement this, without causing any outage?

You would be much better off migrating to RSTP rather than simply configuring uplinkfast. Uplinkfast has been incorporated into RSTP but so have a few other features eg. backbone fast. As long as your switches support it RSTP is the way to go.

As for outages i would always plan an outage when dealing with STP because there can always be unforseen consequences. Promising the business no outage and then having STP reconverge will make you look like you don't know what you are doing. If you are straight with the business they will appreciate it.

As for the migration it is relatively straighforward, please see this document for an example -

PVST+ to rapid-pvst

It is very important with RSTP to -

1) make sure spanning-tree portfast is applied on the correct ports

2) make sure that all inter-switch are running at full duplex - basically RSTP needs to see inter-switch links as P2P links and they need to be full duplex.

Jon

socrates_2000 · ‎09-23-2010

Hi John

Thanks for the explanations. "spanning-tree portfast" has already been configured on the access ports of the 3550 switches to which servers are directly connected. And for interswitch links, i guess they are already full-duplex by default right?

Issuing the command "spanning-tree mode rapid-pvst" on every switches performs the migration right?

Jon Marshall · ‎09-23-2010

socrates_2000 wrote:

Hi John

Thanks for the explanations. "spanning-tree portfast" has already been configured on the access ports of the 3550 switches to which servers are directly connected. And for interswitch links, i guess they are already full-duplex by default right?

Issuing the command "spanning-tree mode rapid-pvst" on every switches performs the migration right?

Yes, they should be full duplex.

"spanning-tree mode rapid-pvst" does indeed perform the migration although if you have uplinkfast/backbonefast configuration you need to remove that config.

But like i say do this during an outage window.

Jon

socrates_2000 · ‎09-23-2010

Hi John

According to cisco technical documentation, the RSTP is backward compatible with STP, however it's mentioned that if the switch running with RSTP detects any other switch with STP, it will make a backward and send STP BPDU messages,and if we enable RSTP in the 2nd switch, the 1st switch will still send STP BPDU till it will be restarted or the STP detection table will be erased ! what are your comments here knowing that in our case we have to enable the RSTP switch by switch ? Do we have to restart the switch 1 in that case? Whats the strategy for migration? Change the mode to RSTP in the 10 access switches first and then the 2 core switches?

Totally we have 12 switches and 28 servers each with 2 ports in one site. How much time it will take for complete migration (convergence time) approximately so that i can ask the customer for a downtime?

We have totaly 3 sites that are in load sharing mode. Do i need to isolate one of the site while doing the migration? Will there be any impact on the other two sites while doing the migration on one site?

IF customer want to implement only uplinkfast in the 10 access switches? Also how much it will take for migration of 10 access switches from pvst to uplinkfast? how much time would take any 3550 switch with uplinkfast enabled, to converge after a possible crash of any 3750 switch? just to compare with RSTP performance (STP take up to 50 seconds) ?

I am sorry for posting lots of question? Am actually from unix background with some cisco knowldge, but the situation is like i have to take care of this.

Jon Marshall · ‎09-23-2010

Hi John

According to cisco technical documentation, the RSTP is backward compatible with STP, however it's mentioned that if the switch running with RSTP detects any other switch with STP, it will make a backward and send STP BPDU messages,and if we enable RSTP in the 2nd switch, the 1st switch will still send STP BPDU till it will be restarted or the STP detection table will be erased ! what are your comments here knowing that in our case we have to enable the RSTP switch by switch ? Do we have to restart the switch 1 in that case? Whats the strategy for migration? Change the mode to RSTP in the 10 access switches first and then the 2 core switches?

Personally i haven't had that issue but if you are taking an outage it should not be a problem reloading switches. If this is an issue then i agree you would be better to do the access-layer switches first then the core switches.

Totally we have 12 switches and 28 servers each with 2 ports in one site. How much time it will take for complete migration (convergence time) approximately so that i can ask the customer for a downtime?

I would ask for an hour per site which gives you a fair bit of leeway if something unexpected happens. Bear in mind the worst case scenario is that you simply end up with STP timers.

We have totaly 3 sites that are in load sharing mode. Do i need to isolate one of the site while doing the migration? Will there be any impact on the other two sites while doing the migration on one site?

How are the sites interconnected ? If they are L3 connected then there is no problem with one site affecting the other. If they are L2 then you can still do one site at a time and all that will happen is that STP timers will be used between sites. One thing to note -

you have 3 switches interconnected sw1 -> sw2 -> sw3 with L2 trunks and 10 vlans. sw1 and sw2 are RSTP capable and sw3 is not. If you have all 10 vlans on all 3 switches then all vlans use standard 802.1d timers. However if you only 2 of the 10 vlans on sw3 and you only have those 2 vlans on sw3 then sw1 and sw2 will use RSTP for the 8 vlans and standard 802.1d timers for just the 2 vlans.

IF customer want to implement only uplinkfast in the 10 access switches? Also how much it will take for migration of 10 access switches from pvst to uplinkfast? how much time would take any 3550 switch with uplinkfast enabled, to converge after a possible crash of any 3750 switch? just to compare with RSTP performance (STP take up to 50 seconds) ?

It's pretty much an immediate switch over to the alternate link with uplinkfast. Again i would take an outage because configuring it alters STP values and this can affect your STP topology.

I don't want to sound overcautious with outages but there are things i am happy to configure on a live production network and there are things that i'm not and STP is definitely one of the things i'm not.

I am sorry for posting lots of question? Am actually from unix background with some cisco knowldge, but the situation is like i have to take care of this.

No problem with all the questions, that's what these forums are for. I started out as a Unix Solaris admin in IT so i know how you end up doing these sort of things

Jon

socrates_2000 · ‎09-23-2010

Hi Jon

The 3 sites are in load sharing mode and redundant pair of 3750 switches in each site are used to connect to other sites using L3 (EIGRP and HSRP).

There are 5 racks/site, Every rack has got 2 3550 switches. Each server has got 2 NIC cards with 2 ports each. There are 2 Vlans, each NIC with two port belongs to one VLAN. One of the port of the NIC is connected to one switch1 and other port to switch2, same way for another NIC. Now all these 3550 have 2 uplinks, one to 3750-1 switch and another to 37500-2 switch.

RSTP

The following is the proceedure i am going follow is the follow:-

1) Make sure all the access ports of 3550 switch that is connected to the server is configured with "portfast"

2) Make sure that, all the trunk ports of 3550 and 3750 switches are operating in Full duplex mode. Or do i have issue the command to force the port as point-to point link, ie "spanning-tree link-type point-to-point" ? Which is the safest option?

3) Then start issue the command "spanning-tree mode rapid-pvst" in global configuration mode in, starting from 3550 access layer switches, and then the two core 3750 switches. If some of the switches are still sending 802.11d BPDUs reload those switches one after the another(i guess i can reload switch2 only after switch1 is reloaded and back into production?)

Please confirm if this proceedure is correct? Usually how long will it take for the RSTP to recalculate parameters and attain stability assuming that there is no topology changes? Anyways am going to request the customer for 1 hour maintenance window.

UPLINKFAST

To enable uplinkfast in all the 10 switches , i have to issue the command "spanning-tree uplinkfast" in global configuration mode right? (or i have to issue the command in interface level?). I have already enabled "portfast" in access ports fo 3550 switches that is connected to the servers. Does Enabling "uplinkfast" requires it to be changed or i leave it alone? I feel it shouldnt be changed, but still i want to make sure.

To implement uplinkfast If i issue the command "spanning-tree uplinkfast" in all 10 switches one after another, how long will it take the network to be stable? Just for curiousity, Does uplinkfast enable on switch affects the same on another switch just like STP?

Most probably customer is going to agree only for UPLINKFAST.

BTW its good to know you were a Solaris admin, I am also from the same domain specializng on SUN hardware and Solaris. But most of the times i am forced to do things on CISCO which am not sure about. Thanks in advance

Thanks

Ajith

Jon Marshall · ‎09-23-2010