Good switches to go for...

[SYNACK]

The problem with 1Gb at the edge is that to utilise it fully you need 10Gb at the core

 

No, to utilize it 'fully' you need semetric bandwidth between the servers and the sum of every single workstations bandwidth. To see a benifit from it you need much less.

 

As I have been saying for the whole thread:

Not every network is the same

you can still get a speedup and noticible benifit during usage without totally semetric bandwidth between the sum of the clients and the servers. Is having that extra capacity good, yes. Is that extra capacity required to see any benifit, no.

[CyberNerd]

Not every network is the same

 

hey, we're starting to agree on something. I've seen a few large 100Mb/s networks trying to run with a 100Mb/s core, they suck. if you only have a few computers it may work for you (earlier you said about 6 and you see the full benefit before you max the core servers). I still fundamentally disagree with your strategy. Networks should be built with scalability and resilience in mind. Putting in 1GB/s edge switches (some recommended on this thread are not even expandable to 10GB/s) without tackling the core first isn't a good plan, it makes the network less resilient - and if you get switches that can't upgrade then it reduces scalability.

[SYNACK]

hey, we're starting to agree on something.

 

We've agreed on this point from the start.

 

I've seen a few large 100Mb/s networks trying to run with a 100Mb/s core, they suck. if you only have a few computers it may work for you (earlier you said about 6 and you see the full benefit before you max the core servers).

Yes they would be very slow, my calcs for 6 were to max the trunk back to the core not the server trunks which are bigger due to multiple catchment areas.

 

I still fundamentally disagree with your strategy. Networks should be built with scalability and resilience in mind. Putting in 1GB/s edge switches (some recommended on this thread are not even expandable to 10GB/s) without tackling the core first isn't a good plan, it makes the network less resilient - and if you get switches that can't upgrade then it reduces scalability.

 

You make it sound like the networks just stop outright and never work again, this is not the case. I do agree that switches with 10GB uplink ability are a good idea for future scaleability but not 100% required. You can push through a 2-4GB trunk which has more redundancy than a single 10GB and can still provide good service. Why spend the whole budget on a use-case that simply does not happen in your environment.

 

Given your assertions I trust you run at least one 10GB link into your servers for each 10 client stations and have at least that many trunks spread to your catchment areas? Or do you simply hobbel all of your hardware investment by limiting it to 10 or 100mbits/s?

[torledo]

Agreed. Thats what I've been trying to say for this entire thread.

 

i'd have to agree with Synack's point, that it's not a hard and fast rule....that 1gbps at the desktop mandates 10gbps in the core.

 

it really has to be down to the particular characteristics of the network traffic in your environment. In an ideal world, resiliant 10gbps links facilitates better future proofing, but for many environments where you don't have the consistent aggregate throughput on core links and to core servers, link aggregated 1gbps links should be more than adequate. The advantage of having the OM3 fibre in place for existing 1gbps fibre links is that with core switches you can swap out 10gbps line cards and have 10gbps top of stack switches at the edge should your traffic patterns indicate a desperate requirement to move to 10gbps.

 

As it is at the moment, core, edge and server 10gbps investments are substantial cost wise. you'd have to do extended monitoring of traffic and usage trends before you could put the business case forward in most schools.

[CyberNerd]

You make it sound like the networks just stop outright and never work again, this is not the case

 

if applications start crashing because a class decides to use all their bandwidth, it sort of does stop working...

 

Given your assertions I trust you run at least one 10GB link into your servers for each 10 client stations and have at least that many trunks spread to your catchment areas? Or do you simply hobbel all of your hardware investment by limiting it to 10 or 100mbits/s?

 

No we can't afford it. Thats one of the reasons I'm saying don't waste money on buying 1GB/s edge switches. We have 1GB/s 'near edge' switches that are 3com 4500G's or 4800G's - they currently have multiple trunked links - so one may have 4 Access points ( 30x 802.11n connections) plus 3 x 4500 switches @2GB/s and then a 2 or 4GB/s uplink to the core. These near edge switches can take 10GB/s modules, as can the 5500G core.

 

When we can afford to replace our fibre optics with OM3 fibre that can carry higher bandwidth and the modules (around £6000 per link) we will do it. When the core gets upgraded (probably to a 7900 series) we'll move the 5500G's out to the near edge and the 4800Gs/4500Gs will get moved to the 1GB/s edge - ie 1GB/s to the desktop. It seems a better plan to me than throwing everything at the desktops and then not having the infrastructure to support things if they actually use it.

[teejay]

We've agreed on this point from the start.

 

Given your assertions I trust you run at least one 10GB link into your servers for each 10 client stations and have at least that many trunks spread to your catchment areas? Or do you simply hobbel all of your hardware investment by limiting it to 10 or 100mbits/s?

 

It's actually around 30 clients at 1Gb going full tilt to saturate a 10Gb link as you won't get 100% full duplex speed due to machine, switch, cabling, operating system overheads and other limitations.

Edited September 28, 2010 by teejay

[SYNACK]

At this point we don't have those kinds of usage patterns to deal with and even with many PCs totally hammering it (saving video) although it did slow down we have not experienced any crashes on our apps because of it. They do 'actually use it' just not to its full extent but as a primary there is generally less load overall. Our network is stable so I don't see why you have an issue with it. We are benifiting from our investment to the full extent that we can with our existing hardware and it works in practice.

 

How about Google Apps as an example, do you throttle all of your stations network cards to 1/(# of workstations)th worth of your avalible internet connection so you can't have contention or do you share all of the avalible link between the stations that are currently accessing it?

 

We don't live in a perfect neat world and with limited budgets/bandwidth and unlimited expectations sometimes things happening in a messy way is better in the end than a sterile algebraic way.

 

It's actually around 30 clients at 1Gb going full tilt to saturate a 10Gb link as you won't get 100% full duplex speed due to machine, switch, cabling, operating system overheads and other limitations.

 

Yes, I was shortcutting it for berevity and dramatic effect, if you see some of my posts eairlier I used the same math as you on my calcs for a 2GB trunk which is where the 3-6 client to saturation number that CyberNerd was using came from.

Edited September 28, 2010 by SYNACK

[teejay]

We do actually throttle per connection internet access, not to internet speed/number of PC's, but to a level whereby having a load of people on iPlayer watching last nights Eastenders won't completely trash everyone elses use of the internet.

As you say though, it's what works for you and your school, the size and structure of the network and your budget.