OSPF is painful

wintermute000

aaron0011 wrote: »

Use or have used following.

Riverbed, BlueCoat, even Cisco WAAS (all WCCP redirects and I don't need to be concerned with the IGP with these devices).

Load Balancers (in our case NetScalers) could give two ****s about the IGP in how we are using them.

Here is the problem right here.

If you want to say, I dunno, pass a CCDP or better still, do design/arch for any other org than your current environment, you better start reading up. You can't very well say 'we're going to use WCCP for our proxy arch because its what I know'. To which a savvy boss would reply 'whats the other options? are any of them cheaper? What architectural implications are there for choosing WCCP vs the other options?'.

I maintain that if you went out there into other environments and topologies you would quickly see how blinkered your current view is. OSPF a huge part of your R&S armory and it is expected for senior engineers/arch/designers to know it backwards.

A large part of your attitude undoubtedly arises from a strongly enterprise centric background, where Cisco rules the shop. Interop at L3 with every other vendor = OSPF or BGP. In SP, Juniper/Brocade is king. In heavy DC / switching environments, Arista and Brocade. In serious load balancing, F5. SBCs? Acme/Sonus. I could go on....

re: your comment about iBGP is interesting... I don't know anybody that uses iBGP as an IGP. I know plenty of SPs or SP-like organisations that use iBGP to overlay a MPLS core in order to provide different VRFs from PEs. You have a big network, yet you want to run iBGP on all your internal routers and deal with a RR nightmare or confed nightmare? Are you aware that two IBGP neighbors that are not directly connected (i.e. intermediate routers not running BGP) will not route properly unless MPLS is employed? Oh and the biggest question: what are you going to use to tie iBGP together? One protocol comes to mind....

just my opinion

aaron0011

wintermute000 wrote: »

If you want to say, I dunno, pass a CCDP or better still, do design/arch for any other org than your current environment, you better start reading up. You can't very well say 'we're going to use WCCP for our proxy arch because its what I know'. To which a savvy boss would reply 'whats the other options? are any of them cheaper? What architectural implications are there for choosing WCCP vs the other options?'.

For web filters, proxies, WAN accelerators, and the like...there isn't a better way of doing it than WCCP. So the counter of what's another way by your boss isn't a real world scenario as far I am concerned. A tech savvy manager in an enterprise environment would want the best practice methods in the network...and if he doesn't I don't want to support said network. Back to WCCP, L2 redirects are ideal because of less overhead but even at L3 you're dealing with a select number of devices in a service group transversing a GRE tunnel. WCCP is of course a Cisco protocol but thankfully a ton of devices support it.

wintermute000 wrote: »

A large part of your attitude undoubtedly arises from a strongly enterprise centric background, where Cisco rules the shop. Interop at L3 with every other vendor = OSPF or BGP. In SP, Juniper/Brocade is king. In heavy DC / switching environments, Arista and Brocade. In serious load balancing, F5. SBCs? Acme/Sonus. I could go on....

This is absolutely correct in my case. I work in an enterprise and not a SP. If it were the other way around I know my experiences would be different. Cisco doesn't make the best of everything but they do make a wide range of product that falls under the same very solid support model. In the enterprise world, great support when needed is absolutely critical.

DevilWAH

I have to say the chances are you will have to use a non EIGRP protocol at some point in your career, especially if you want to include non traditional routers like firewalls and loadbalancers in to your network. Take the likes of firewalls, Cisco does not make the best firewalls on the planet, yes they are not bad, but if you are in a business that has a strong security element you can be sure they will be using other stuff and unless you want a lot of static routing you want to run OSPF so you can dynamically include them in your routing solutions. Even in a CISCO house your not going to chose CISCO for some areas because even if you have an unlimited budget there are much better solutions for some things out there. So if you don't have a decent understanding of OSPF then you cut you self of from a huge chunk of the networking jobs. Even if you don't know it in it's completeness you need to know how to do a basic set up and how to integrate them in to the core if that is running EIGRP.

But more than that, OSPF was developed in the 1980 while EIGRP didn't come along till the 90's, you might say well that just goes to show its the better protocol and to be sure its easy to configure and if I had the choice I would go for it every time. But OSPF was developed when the hardware did not have the proformace it does now, it was developed to support the rapidly expanding networks at the time and all the area stuff and the LSA types where a way to allow low spec routers to deal with large expanding network. EIGRP kind of glosses over this by taking care of it all for you, any fool can set up EIGRP with out knowing any thing about routing and this in its self is dangerous. OSPF on the other hand I think is a much better tool when you are learning to actually make you step back and think through how things are working. In your terms "ERGRP is a lazy protocol" its a breeze to set up, and honestly you really only need 3 or 4 commands to do it. But that plug and play can come back to haunt a engineer later, as things grow and grow and you end up with a poorly thought out and cumbersome network. OSPF on the other hand right from the start teaches you about how to build up a network it makes you plan much better. You kind of set up OSPF how you would design a good network, not just throw every thing in one bucket and forget about it.

And as for the iBGP idea, iBGP is not for internal routing in an enterprise network, its for joining edge BGP routers together so they can exchange information and router sent from eBGP neighbors across the local AS. as mentioned iBGP needs an IGP to run on top of, so you back to looking at either EIGP or OSPF for that.

I have worked with a stack load of Bank and Oil companies as clients when I worked in a CiSCO solution and services provider (one of the largest in the world). I would say 60 - 70% ran the majority of the routing using OSPF, and even those that had EIGRP as the main protocol had many areas of OSPF for specific reasons. In the public sector or MD EIGRP is often out of the picture as you cant use propitiatory protocols.

The thing is once you accept that in some cases even if its only in a tiny proportion of a network you need to run OSPF such as on your firewalls. Then whats the point of implementing EIGRP at all and having to run 2 protocols? Once learnt OSPF is no worse than EIGRP for 99.9% of cases, you can tune it and its not hard to set up once you get your head round it. A well designed OSPF network is more intuitive to look at and trouble shoot and takes no longer to configure. So if I had the luxury of a 100% cisco house and no reason to use OSPF at all then I would go EIGRP 100%, but if I have to use OSPF in one place then I am going to run it across the shop and not worry about having to manage two protocols or deal with redistribution's issues. I know of one network that ran 100% eigrp, that was the first one I ever set up, since then its either been 100% OSPF or a mixture.

wintermute000

aaron0011 wrote: »

For web filters, proxies, WAN accelerators, and the like...there isn't a better way of doing it than WCCP. So the counter of what's another way by your boss isn't a real world scenario as far I am concerned. A tech savvy manager in an enterprise environment would want the best practice methods in the network...and if he doesn't I don't want to support said network. Back to WCCP, L2 redirects are ideal because of less overhead but even at L3 you're dealing with a select number of devices in a service group transversing a GRE tunnel. WCCP is of course a Cisco protocol but thankfully a ton of devices support it.


This is absolutely correct in my case. I work in an enterprise and not a SP. If it were the other way around I know my experiences would be different. Cisco doesn't make the best of everything but they do make a wide range of product that falls under the same very solid support model. In the enterprise world, great support when needed is absolutely critical.

This is presumably why most big Riverbed deployments (best in class lets not forget!) goes inline? Anyhow arguing about specific WCCP topologies is a furphy and you know it, or are you seriously saying that vendor L3 interop doesn't matter because proxies/LBs/accelerators can use WCCP!?!?!?!?!? (and of course this is not only contentious, but also assumes WCCP will always be the best option assuming we agree that it is the best option... which patently in many cases it is not, but I digress)

A truly savvy manager in ANY network goes for best bang for the buck. Best practice may not be the best option if it spends more money for gains that don't really matter. Its all about ROI. I can't remember the last time I sat in an Enterprise project where the design options were broken down on a ROI basis and the best option picked. Its always 'here is the way we're going to do it - Cisco said its the best. I read it here in this white paper'. Enterprise is HORRIFIC for this. When was the last time you picked a design because you actively investigated, labbed and eliminated the alternatives are more costly / delivering insufficient capability vs the spend incurred? (in a documented manner all tabled in writing and signed off, not a 5 minute chat in the corridor between the lead techies)

Eliminating other vendors from your arsenal - and eliminating valid design options because of your personal dislike for XYZ protocol - makes you a more limited designer/arch. Regardless of enterprise or SP. What would you say if I told you Brocade sells 1RU 24x1G port devices that run full MPLS, full BGP table and costs HALF that of an ASR1001 (that has 4 measley ports)? But to put that baby in you need OSPF or ISIS as the underlying IGP?

A few of your comments indicate to me (only my opinion) that perhaps there are things you haven't seen or encountered that if you did, would open your eyes. Put it this way, its immediately obvious to me that you are not a R&S specialist - not a bad thing and I don't pretend to know what exactly you do - but it IS a bad thing if you are going to make some of the assertions you make e.g. 'jus run IBGP internally', your comments re: OSPF, blanket 'WCCP is the best, always and ever' (really?!?!??!) etc.

I apologise in advance if any of the above is a bit aggressive, just my opinion.

aaron0011

wintermute000 wrote: »

This is presumably why most big Riverbed deployments (best in class lets not forget!) goes inline? Anyhow arguing about specific WCCP topologies is a furphy and you know it, or are you seriously saying that vendor L3 interop doesn't matter because proxies/LBs/accelerators can use WCCP!?!?!?!?!? (and of course this is not only contentious, but also assumes WCCP will always be the best option assuming we agree that it is the best option... which patently in many cases it is not, but I digress)

A truly savvy manager in ANY network goes for best bang for the buck. Best practice may not be the best option if it spends more money for gains that don't really matter. Its all about ROI. I can't remember the last time I sat in an Enterprise project where the design options were broken down on a ROI basis and the best option picked. Its always 'here is the way we're going to do it - Cisco said its the best. I read it here in this white paper'. Enterprise is HORRIFIC for this. When was the last time you picked a design because you actively investigated, labbed and eliminated the alternatives are more costly / delivering insufficient capability vs the spend incurred? (in a documented manner all tabled in writing and signed off, not a 5 minute chat in the corridor between the lead techies)

Eliminating other vendors from your arsenal - and eliminating valid design options because of your personal dislike for XYZ protocol - makes you a more limited designer/arch. Regardless of enterprise or SP. What would you say if I told you Brocade sells 1RU 24x1G port devices that run full MPLS, full BGP table and costs HALF that of an ASR1001 (that has 4 measley ports)? But to put that baby in you need OSPF or ISIS as the underlying IGP?

A few of your comments indicate to me (only my opinion) that perhaps there are things you haven't seen or encountered that if you did, would open your eyes. Put it this way, its immediately obvious to me that you are not a R&S specialist - not a bad thing and I don't pretend to know what exactly you do - but it IS a bad thing if you are going to make some of the assertions you make e.g. 'jus run IBGP internally', your comments re: OSPF, blanket 'WCCP is the best, always and ever' (really?!?!??!) etc.

I apologise in advance if any of the above is a bit aggressive, just my opinion.

Riverbed recommends inline a lot of time because they don't understand WCCP. I've got tons of Riverbed experience and worked with their SEs a lot over the years so this statement isn't just something I made up.

Will comment more later. Busy busy day today. No hard feelings at all wintermute. Good discussion.

aaron0011

wintermute000 wrote: »

A few of your comments indicate to me (only my opinion) that perhaps there are things you haven't seen or encountered that if you did, would open your eyes. Put it this way, its immediately obvious to me that you are not a R&S specialist - not a bad thing and I don't pretend to know what exactly you do - but it IS a bad thing if you are going to make some of the assertions you make e.g. 'jus run IBGP internally', your comments re: OSPF, blanket 'WCCP is the best, always and ever' (really?!?!??!) etc.

I am definitely not a Routing Specialist (never made such a claim) and I don't think I want to be. My expertise lies in Voice and due to my experience with Switching I was able to knock out SWITCH at a quicker than normal pace. I'm also doing a lot of UCS, VMware, Nexus 1000V these days because of prior VMware experience and the convergence of that area with Voice. I really enjoy that work more than learning about OSPF or any routing protocol really.

FWIW, I am not taking offense to your comments here. I read and post on this forum because it's fun to share knowledge and experiences with others in the field. You hold a CCNP Voice yet earlier in the thread suggested let's stick with MGCP and not mess with SIP. Any decent Voice engineer working in today's environments would never say that. Maybe it was in jest so I didn't feel the need to hammer you on it.

My OP isn't the mind set to have...of course I realize that. It was just my first reaction after really diving into OSPF passed the CCNA level. My point here is I work in an enterprise where its not used and will not be for the foreseeable future. Maybe one day that will change but in this particular network EIGRP works well. No need to change something that works. I will reinterrate again and again that I can learn every detail about OSPF needed for the ROUTE exam and lab GNS3 forever but if I am not using it every day, a true understanding and passion for OSPF won't be there.

I'm looking forward to your examples on handling WAN accelerators and web filter/proxies with a better solution than WCCP. If there is a better way I can join groups of devices into a service group, load balance said devices, and control traffic flows all the way to Layer 7 I am all ears. OSPF won't be the solution to that one, I am certain. And again, Riverbed recommended inline is a borderline joke. If you want to cable multiple appliances with large amount of ports inline all over your Data Center in between your Core and WAN routers and then using something else to Load Balance then knock yourself out.

wintermute000

aaron0011 wrote: »

I am definitely not a Routing Specialist (never made such a claim) and I don't think I want to be. My expertise lies in Voice and due to my experience with Switching I was able to knock out SWITCH at a quicker than normal pace. I'm also doing a lot of UCS, VMware, Nexus 1000V these days because of prior VMware experience and the convergence of that area with Voice. I really enjoy that work more than learning about OSPF or any routing protocol really.

FWIW, I am not taking offense to your comments here. I read and post on this forum because it's fun to share knowledge and experiences with others in the field. You hold a CCNP Voice yet earlier in the thread suggested let's stick with MGCP and not mess with SIP. Any decent Voice engineer working in today's environments would never say that. Maybe it was in jest so I didn't feel the need to hammer you on it.

My OP isn't the mind set to have...of course I realize that. It was just my first reaction after really diving into OSPF passed the CCNA level. My point here is I work in an enterprise where its not used and will not be for the foreseeable future. Maybe one day that will change but in this particular network EIGRP works well. No need to change something that works. I will reinterrate again and again that I can learn every detail about OSPF needed for the ROUTE exam and lab GNS3 forever but if I am not using it every day, a true understanding and passion for OSPF won't be there.

I'm looking forward to your examples on handling WAN accelerators and web filter/proxies with a better solution than WCCP. If there is a better way I can join groups of devices into a service group, load balance said devices, and control traffic flows all the way to Layer 7 I am all ears. OSPF won't be the solution to that one, I am certain. And again, Riverbed recommended inline is a borderline joke. If you want to cable multiple appliances with large amount of ports inline all over your Data Center in between your Core and WAN routers and then using something else to Load Balance then knock yourself out.

You got me completely wrong. I said RHETORICAL question. It was to illustrate what I consider the absurdity of what you're saying.
Given that I've done ISDN to CUBE migrations till the cows come home, and then ripped out CUBEs for real SBCs for a carrier, I would be last person in the world to advocate MGCP over SIP LOL

aaron0011

wintermute000 wrote: »

You got me completely wrong. I said RHETORICAL question. It was to illustrate what I consider the absurdity of what you're saying.
Given that I've done ISDN to CUBE migrations till the cows come home, and then ripped out CUBEs for real SBCs for a carrier, I would be last person in the world to advocate MGCP over SIP LOL

I missed the rhetorical part. I see your point.

tink334

Learn the stuff and forget it. You will alway pick up on it quickly if you study it.

aaron0011

tink334 wrote: »

Learn the stuff and forget it. You will alway pick up on it quickly if you study it.

Exactly what I am going to do. If I ever do an OSPF project I'll be going over all the details again anyway based on requirements. I'll never be the go to guy routing expert and I am okay with that. Data Center and Voice are much more fun IMO.

I plan to post my passed ROUTE thread by April 1st. I've been reading the FLG and watching Nuggets off and on for the past couple of months. Time to really hit OSPF hard and lab lab lab in GN3.