Well in my knowledge of etherchannels spanning-tree treats a port-channel as a single link. So when one link goes down the other port keeps forwarding. Now since your links aren't fully meshed I think you will always have the problem of losing packets. To get full redundancy no matter what happens I would fully mesh etherchannels on all switches and use HSRP for redundant gateway.
> Date: Thu, 10 Mar 2011 08:10:33 -0600 > Subject: Re: [OSL | CCIE_RS] Rapid Spanning Tree convergence times > From: [email protected] > To: [email protected] > CC: [email protected]; [email protected] > > Just to add a little more to this, even after tuning all the rstp > timers down to their minimum I don't see any noticeable change. I > still drop 1 packet between convergence. So I guess it's better to > keep the default timers to keep cpu usage down if it doesn't really > gain me anything to go lower. Anyone have an opinion on this? > > > On Wed, Mar 9, 2011 at 8:53 PM, marc abel <[email protected]> wrote: > > I'm steadily improving the network from little/no redundancy to > > complete redundancy. I have to pick my budget battles right now and > > frankly there are a few other places I'd rather improve first. > > Fortunately all my server switches are linked via ether-channel. > > Although no matter how many links in your bundle, if you lose your STP > > root, then you are going to have a re-convergence. I had a core switch > > roll over on me due to a "parity error" (according to TAC) a few > > months ago. Things recovered really well, but I haven't implemented > > the voice yet so I didn't have anything quite that latency dependent > > at the time. > > > > On Wed, Mar 9, 2011 at 8:38 PM, Jay Taylor <[email protected]> wrote: > >> Any chance of just doubling your links for each path? That would only > >> require 2 extra pairs of fiber to each IDF rather than uplinking each > >> switch. But then again, that 1 ping missed with RSTP was with default > >> timers > >> and I'm sure you could trim that down. > >> > >> > >> On Wed, Mar 9, 2011 at 9:32 PM, Michael Smith <[email protected]> wrote: > >>> > >>> Well you guys are right. I've always dealt with switches that have been > >>> etherchannelled to another switch so I never really dealt with switches > >>> that > >>> are single linked like that. I just don't know the reason why anybody > >>> wouldn't etherchannel their switches together. Hey I guess you live and > >>> you > >>> learn! > >>> > >>> > >>> > >>> > Date: Wed, 9 Mar 2011 20:28:21 -0600 > >>> > Subject: Re: [OSL | CCIE_RS] Rapid Spanning Tree convergence times > >>> > From: [email protected] > >>> > To: [email protected] > >>> > CC: [email protected]; [email protected] > >>> > > >>> > Right because in my situation one of the links to the cores is going > >>> > to be in blocking state. I don't see any way around that. > >>> > > >>> > On Wed, Mar 9, 2011 at 8:25 PM, Jay Taylor <[email protected]> wrote: > >>> > > Even with HSRP and sub-second hellos you could lose pings depending on > >>> > > how > >>> > > STP needed to converge. > >>> > > > >>> > > On Wed, Mar 9, 2011 at 9:24 PM, Michael Smith <[email protected]> > >>> > > wrote: > >>> > >> > >>> > >> Well I'm talking as far as the VOIP phones go. They obviously need a > >>> > >> gateway and to not miss any pings you can always turn on HSRP. > >>> > >> I'm not saying HSRP has anything to do with spanning tree. Just > >>> > >> thinking > >>> > >> about the fact of not losing any pings. > >>> > >> > >>> > >> > >>> > >> > >>> > >> ________________________________ > >>> > >> Date: Wed, 9 Mar 2011 21:05:04 -0500 > >>> > >> Subject: Re: [OSL | CCIE_RS] Rapid Spanning Tree convergence times > >>> > >> From: [email protected] > >>> > >> To: [email protected] > >>> > >> CC: [email protected]; [email protected] > >>> > >> > >>> > >> Maybe I'm missing something but how can HSRP (or first hop redundancy > >>> > >> protocol) replace STP/Etherchannel? Even if 2 of the Catalyst > >>> > >> switches in > >>> > >> that topology were L3 gateways and ran HSRP you still need to deal > >>> > >> with the > >>> > >> L2 loop that exists. > >>> > >> > >>> > >> > >>> > >> On Wed, Mar 9, 2011 at 8:50 PM, [email protected] > >>> > >> <[email protected]> > >>> > >> wrote: > >>> > >> > >>> > >> Well if yor timers are that bad for VOIP you can always use hsrp if > >>> > >> you > >>> > >> don't want to use the etherchannel option. You can tune hsrp down to > >>> > >> milliseconds if you wanted to. Of course your distribution switches > >>> > >> need to > >>> > >> support an enhanced IOS image > >>> > >> > >>> > >> Sent from my Verizon Wireless Phone > >>> > >> > >>> > >> ----- Reply message ----- > >>> > >> From: "Jay Taylor" <[email protected]> > >>> > >> Date: Wed, Mar 9, 2011 7:57 pm > >>> > >> Subject: [OSL | CCIE_RS] Rapid Spanning Tree convergence times > >>> > >> To: "marc abel" <[email protected]> > >>> > >> Cc: <[email protected]> > >>> > >> > >>> > >> > >>> > >> Enable portfast on the host ports and you'll see a much quicker > >>> > >> transition. > >>> > >> Just labbed this up and with portfast enabled I lost a single ping > >>> > >> during > >>> > >> the failover. Without it enabled I lost 12. > >>> > >> > >>> > >> For the VoIP question - in production I'd recommend building with > >>> > >> Etherchannels just so STP never needs to converge. > >>> > >> > >>> > >> > >>> > >> On Wed, Mar 9, 2011 at 5:41 PM, marc abel <[email protected]> wrote: > >>> > >> > >>> > >> > I have 4 switches connected in a loop. > >>> > >> > > >>> > >> > Cat1-------------Cat2 > >>> > >> > | | > >>> > >> > | | > >>> > >> > Cat3----------Cat4 > >>> > >> > > >>> > >> > > >>> > >> > Cat 1 is the root, Cat 2 is the secondary root. All the switches > >>> > >> > are > >>> > >> > set to RPVSTP and I have confirmed that show spanning-tree shows > >>> > >> > RSTP > >>> > >> > as the protocol. Cat 4 shows it's interface to cat3 as it's root > >>> > >> > port > >>> > >> > and the interface to Cat3 as the Alternate. I have not tuned any > >>> > >> > timers. > >>> > >> > > >>> > >> > What should be the convergence time in this situation? > >>> > >> > > >>> > >> > If I run a ping from a host attached to Cat4 to a host attached to > >>> > >> > Cat1 and then I shut the Cat1-Cat3 interface (on the Cat1 side) it > >>> > >> > takes about 32 seconds before pings pick back up. I thought RSTP > >>> > >> > was > >>> > >> > supposed to converge in about 6 seconds? > >>> > >> > > >>> > >> > Another question, what is the fastest recovery time we can tune > >>> > >> > down > >>> > >> > to from RSTP? How do others tune this for VOIP? I know that I can > >>> > >> > get > >>> > >> > sub second convergence from OSPF but not all my switches have an > >>> > >> > appropriate image to run ospf. > >>> > >> > > >>> > >> > > >>> > >> > Thanks in advance. > >>> > >> > > >>> > >> > Marc > >>> > >> > _______________________________________________ > >>> > >> > For more information regarding industry leading CCIE Lab training, > >>> > >> > please > >>> > >> > visit www.ipexpert.com > >>> > >> > > >>> > >> _______________________________________________ > >>> > >> For more information regarding industry leading CCIE Lab training, > >>> > >> please > >>> > >> visit www.ipexpert.com > >>> > >> > >>> > >> > >>> > >> > >>> > >> > >>> > > > >>> > > > >> > >> > > _______________________________________________ For more information regarding industry leading CCIE Lab training, please visit www.ipexpert.com
