Network Scalability

[richard.m.gilbert]

Hello all,

The term Network Scalability is tossed around a lot but it seems a bit vague to me. What does it mean to 'scale a network?'

Thanks,

Richard

[Jon Marshall]

Richard

You may get a variety of answers but to me a scalable network is a network that can cope with the existing demands placed upon it but also one that be expanded to meet future demands in a planned, graceful way.

When you scale a network you design for the current needs but you also design with future needs in mind. If demand increases in your network by for example 40% over the next 6 months you should not need to redesign the network from scratch.

Hence the reason for Cisco's 3 tier architecture core/distribution/access. Each tier has a specific purpose and if designed correctly you can increase capacity in one without having to redesign other parts.

Jon

[scottmac]

I agree with Jon and want to emphasize the words "planned and graceful."

You can expand any network, expanding it "easily" and without causing some other part of the network to suffer is what planning is all about.

For example: you *can* daisy-chain switches (switchport-to-switchport, versus a cascade/cluster port)to gain distance and port count ... but the scalability is limited (and it's bad design); eventually the links closest to the core are overloaded, even with nominal traffic on the downstream switches because of the cumulative traffic from downstream.

"Scalable" implies (relatively) painless expansion with no decrease in functionality, and no significant re-engineering of the existing infrastructure.

Scott

[wbenton-0]

To "scale a network" means to "grow a network" or "increase the size of a network".

Thus scalability means the ability to be able to scale or adapt the network for future growth.

On another note, limited scalability means that there are limitations placed on how large a network can be grown or increased. Those limitations could be bandwidth, number and or types of nodes, protocols, VLANs or perhaps any other limiting factor which limits it's scalability to a maximum of {fill_in_the_blank}!

Walt

[richard.m.gilbert]

Thanks Jon and Walt ... I use to only think in terms of IPs but I guess it is far more than that. Port density, router memory, router cpu, as well as IP density and summarization plan. But when the BSCI textbook states that an advantage of EIGRP is that it is scalable in terms of hardware resources and network capacity ... what are they meaning? Is that simply a fancy way of stating that they don't send out the entire routing table to their neighbors every X seconds as compared to another DV protocol like RIP?

Richard

[Jon Marshall]

Richard

"Is that simply a fancy way of stating that they don't send out the entire routing table to their neighbors every X seconds as compared to another DV protocol like RIP?"

Pretty much yes. Once an initial adjacency has been formed and full routing updates exchanged from then any routing change only generates a partial update. Hello's which have minimal impact on the network are used as keepalives.

Eigrp also has other efficiences such as the Feasible Successor concept and you can also configure a maximum percentage of the bandwidth that EIGRP is allowed to use which can be useful on very slow links.

From experience though, EIGRP is scalable but only really when used in conjunction with IP summarisation. Without summarisation a link down in one part of your network has a knock on effect to the rest of the network - assuming no feasible successor.

Compare this with OSPF which has an inbuilt hierarchy due to it's area concept.

Jon

[richard.m.gilbert]

Jon,

It has me a bit confused because I would think that every protocol is more scalable than other protocols in certain areas. I would think that of them all, OSPF is the most scalable protocol because it can be broken up into areas.

Richard

[Jon Marshall]

Richard

Yes that was what i was getting at ie. because of it's area concept OSPF is a very scalable IGP - apologies if i gave a different impression.

One thing to bear in mind with OSPF is that it can be quite demanding on router resources eg. CPU/memory so as you say they all have their pros and cons.

Jon

[richard.m.gilbert]

Jon,

No wrong impression took ... I think I was aggresively agreeing with you :). I appreciate the candid answers as I learn a lot from them. I think I see scalability in a more broader sense with a better understanding. I think it contains things I thought about before but never really put it as network scalability. I do now though.

Richard

[Jon Marshall]

Richard

No problem. I think that's one of the best things about NetPro in that it is focussed on solving everyday problems as well as theory and there are some very smart people who hang around these forums who have a vast amount of knowledge as well as experience.

Jon

[richard.m.gilbert]

Jon,

I totally agree. This forum is an incredible resource. The people here really understand the value in answering the questions because that is one of the best ways to learn ... as well as asking questions. Going through the books in an attempt to study can be difficult because it is easy to take a collection of data and go in different intended directions. The result is not getting the concept needed to pass the test ... though many things are learned anyways. Studying for the BSCI is much easier with my router/switch lab, the BSCI books, and NetPro. Without NetPro, understanding can be a theory instead of a fact ... if that makes sense.

Richard

[wbenton-0]

Many may claim that EIGRP is advantageous over OSPF because of it's fast convergence, but when your network scales larger and larger, you start running into the limitations of EIGRP regardless of it's fast convergence.

The following is a quote from a Network World article (http://www.networkworld.com/community/node/16276) which I highly recommend you read. There is no need in me re-explaining what this article does a very good job of.

>>>The primary scaling limitation with EIGRP is that it doesn't have a capability for setting internal boundaries, important for controlling prefix summarization and database sizes, the way OSPF areas do. You can artificially do this by using multiple EIGRP processes, but why use a kludge to accomplish something OSPF does as an integral part of the protocol?<<<

On another note, scalability can also mean the adaption or use of new protocols or protocols which you have yet to use.

Just like you can scale an EIGRP network to an OSFP network, you can also scale your non-VoIP network to support VoIP and/or wireless even if you don't currently use those now. But you have to have the equipment in place which can scale to those new or previously unused protocols.

Likewise, you can scale CDP (Cisco Discovery Protocol) which is a Cisco proprietary protocol to LLDP (Link Layer Discovery Protocol) and inter-operate with non-Cisco devices as well.

Thus scaling and/or scalability has a very wide meaning.

Walt