Sunday, May 4, 2014

Time To Get Rid Of 7 IP Networking Myths - Based On Recent Interviews I Did





In last 6 months I got opportunity to work closely with HR teams from different regions within company and must have interviewed well over 200 People which gave me good opportunity to talk to so many candidates ranging from CCNA to CCIE & even Multiple CCIEs.

During this whole process I came across couple of Myths related to IP Networking which I would want to share so that together we can spread more awareness among other Engineers & Clients.


Myth 1 : vPC eliminates STP into the Network.

Well vPC tweaks STP behavior quite a bit for it’s operation but STP is still running in backend as fail safe. Even more vPC Primary switch send regular STP BPDUs by default over vPC member ports. Well the only solution to get rid of STP completely today from Cisco’s perspective is implementing Fabric Path.


Myth 2 : EIGRP is a better protocol for Cisco only network since it has better convergence features such as Feasible Successor if compare to OSPF. Same thought people have when comparing ISIS with OSPF.

Well convergence in itself is domain now from IP Networking perspective or perhaps always was. It’s just people these days talk more about it as requirements have grown tight in last 5 years because of Next Generation Network Requirements and more sensitive applications.

Most routing protocols don’t spend too much time for re-calculation once the primary path gets failed. The most contributing factor in convergence is detection of failure rather. A simple example would be to have two EIGRP neighbors connected through multiple links including a Ethernet connection through L2 Switch which is currently primary path. If link between L2 switch & Router 2 fails, Router 1 will notice the failure only after dead timer expire and than will run DUAL for recalculation depending upon it has Feasible successor or not. So in this case it’s not going to be any different than OSPF or IS-IS.


Myth 3 : OTV is the best solution for VMotion over Data Center Interconnect (DCI) or any Layer 2 extension requirements.

Well It depends. The part of equation here is VMotion requirements itself. For Hot VMotion the end to end delay requirements is less than or equal to 5 msec which in new ESXi version got relaxed to 10 msec I guess. So if the delay is higher and you have Hot VMotion requirements, OTV won’t do any good as DCI. Also choosing DCI Technology Between OTV and others like MPLS Martini Style, VPLS etc has lot to do with Scale related to number of MAC Addresses to be learned & Exchanged.


Myth 4 : One major benefit of using MPLS is that Router Switches packets faster since it uses LIB


Well it use be true in history but ever since the introduction of Hardware based forwarding in platforms and platforms having great amount of CPU and Memory, it’s not true any longer.


Myth 5 : Nexus 2000 is a Switch

Well it can’t perform local switching between 2 ports which might be even in Same VLAN. All the switching is still performed by Parent Switch (7k or 5k) based on VN Tags.


Myth 6 – Once the EIGRP SIA Timer Expires, The Neighbors are bounced

Well Cisco solved this problem I guess almost 10 years back with implementation of SIA Query & Replies. It’s just of those things which are not documented in most Cisco Press Books J covering EIGRP


Myth 7 – Traceroute is part of ICMP Suite

Well traceroute is an application but not a protocol or part of any protocol suite in itself. Traceroute has different implementations using ICMP, TCP & UDP by different vendors.


HTH...
Deepak Arora
Evil CCIE

2 comments:

Andrew said...

Hi,
Myth 2 :
.....So in this case it’s not going to be any different than OSPF or IS-IS.

That about IP LFA FRR extension for ospf or is-is?

Deepak Arora said...

Hi Andrew,

Though LFA and RLFA under IP FRR will certainly help in terms of overall network convergence depending upon topology we have. Idea here is the important piece missing is Network Failure Detection. No matter how much I have reacted upon in order to faster convergence but unless I detect failure I can't respond which is the case here under Myth 2.

To faster the detection process there are several techniques like BFD, IP SLA, Having point to point connections without L1 or L2 bumps in between, Reducing career delay to zero etc.

HTH...
Deepak Arora