Hi Farinha,

I would still argue that any sizable organization should be deploying 1Gbps to the desktop. As for my current employer we’re a little further along than when I wrote this article but I would guess (I should probably run the numbers now that you’ve peaked my curiosity) that we’re probably 40/60 with 40% or the switch ports being 1Gbps and the remaining 60% being 100Mbps (it takes time to upgrade 37,000 ports). The applications (in my honest opinion) are only getting worse. I have seen a few recent true web based (very heavy Javascript) applications that are very well written – but only because they need to be when they are run across the public Internet.

We’ve been all digital for a few years now from an imaging perspective. It can be hard pleasing those radiologists! ;)

Cheers!

Do you still have the same opinion?

We are facing a similar discussion for a medical imaging department… Though the application os quite inefficient, it is a fact that 1 Gbps access with a recent workstation will reduce the transfer of medical images quite a bit…

But should we thrive for the apps guys to optimize their apps or just give them what they want (bandwidth)?

As mentioned I think the circumstances are different for almost everyone hence the broad array of responses. In my organization we’ve been 100Mbps switched to the desktop since about 2007 and we’re pushing a lot of rather ugly and inefficient client server applications where application performance (and acceptability) is directly tied to link speed. While people frown on the old line, “just throw bandwidth at it”, this solution still applies today where the people selecting the applications and signing the contracts aren’t really interested in the challenges that IT will face trying to implement their monstrosity. So the solution more times than not is to upgrade the edge switch to 1000Mbps because “it’s a network problem”.

@Jon, I wasn’t that comfortable myself reading that statement… thanks for pointing out the flaw.

Just one follow-up to @Thomas’ comment regarding link utilization and graphing (MRTG/Cacti). You do need to recognize that there is a big difference between 1 minute polling interval and 5 min polling interval. You might have an application that fills the 100Mbps pipe for 10 seconds but on a (5 minute average) graph it might not even show up as more than a little blip. So in essence you think that the link is only 2% utilized, however, it might be 100% utilized dozens of times a minute but the average overall utilization over a 5 minute period only accounts for 2% utilization. I have a number of interfaces that we monitor on both 1 minute and 5 minute intervals and the difference can be dramatic.

Cheers!

I would like to add two remarks and post my personal opinion:

1) The hidden cost: power consumption. Gbps is 60% more power hungry than 100Mbps, even when in idle.
At both sides of the line…
(actually the Avaya Power Saver sets Gbps-ports to 100Mpbs when activated)
More power consumption results in higher cooling costs.

2) It’s easy to be fooled when using MRTG/Cacti/… to monitor link usage. These poll every 5min, so flatten out the peaks. One download taking 10 at 1Gbps vs. 1min at 100Mbps will show exactly the same. Though users will definitely notice a speed difference. Then we are talking radiologist, HD Video, CAD, DTP, scanning/archiving, …
Normal desktop, phone, thin client, AP almost will not feel the 1Gbps advantage.

Currently I would make following choice (for bigger networks):
new switches = full Gbps and PoE.
Depending on needs: set ports to 100Mbps autoneg. maximum.
Initial CapEx is higher, but results in uniform network, that can provide all required connectivity where needed.
OpEx will definitely be lower in the long term because of simplicity & flexibility.

Try calculating cost per day for lifetime, and cost of swapping to 1Gbps-switches in near future, with some probability factor.

KR,

Thomas

Depending on your environment, check your current uplink utilisation to see if you need to upgrade to 10 Gigabit or just add more Gigabit links to your MLT uplinks? Even though your users are only at 100Mb/s you should still be able to plan what the expected utilisation will be when migrating to 1Gb/s at the edge?

We currently have the standard 2 x 1Gb/s MLT uplinks from the backbone to the edge for 100Mb/s client connectivity but we’re wondering if we’d also need to increase the uplinks when moving to 1Gb/s to the clients? 2 x 10Gb/s in the foreseeable future maybe when the equipment becomes cheaper (XFP, SFP+, …)?

Our ESX farm would have to be migrated to 2 x 10Gb/s uplinks first ofcourse ;-).

Technically, no. The propagation delay, the time it takes for the electrical pulses to travel from one end to the other is the same regardless of the bandwidth. It is a function of physics and is roughly equated to the speed of light.

Their comment, however, is true when you consider the “wire” to include the serialization delay. The rate at which data is placed on the wire (serialization) does depend on he bandwidth. Effectively, the serialization and propagation delays together as a time are less as the bandwidth increase. The gains, however, are from the decreased serialization.

I can’t see us ever using Gigabit on the edge unless the price is comparable with 100Mbps?

We wring every last dollar out of our hardware and I can see the ERS4548GT’s being used for at least another 6-8 years.

Now if I could just make the Avaya Tech Support web-site run faster so downloads of software updates don’t take so darn long. What a dog…

For the Gig to the desktop debate, I think the answer is “it depends”. For our 1,300 node network, 100Mbps is king, mostly. What we are finding is that one of our apps is very packet hungry. The problem isn’t the amount of data, but the inefficiency of the application and it’s need for a lot of data in very small sizes. And since the vendor says “it works fine for us”, obviously it must be our network, right? So, what we found is if we put these nodes on a gig port, the app worked “just like” the vendor expected.

I would agree that most applications, barring GIS/CAD, Graphic Art/Video, and poorly designed APPs, need only 100Mbps.

One argument that I have heard from others, is that GigE is more efficient overall, because the packet, no matter what size, is on the “wire” for a shorter amount of time. I personally think this is a good argument for uplinks back to the Core and for all verticals, but for the desktop, I think that is only applicable to real-time apps that would saturate a 10/100 Mbps connection.

I am curious to what others have to say. Sean

Everbody wants today gigabit ethernet to the destop.
But if you take a deeper look at the traffic that is flowing on Desktop Ports you will find only a very few ports that really has a high traffic utilization that requires gigabit ethernet.