Gigabit ethernet

[jteef]

I am looking for a fast home network. What is a relatively low cost implementation I could do with gigabit ethernet on copper?

This isn't much of a practical application. A 100mb/s network would be enough for my application, but that isn't the point is it?

so basically what I am looking for are gigabit switches or hubs that don't cost too much money. Is there anything out there for less than $500? The netgear GS504T is about $700 on pricewatch. Is there anything cheaper?

thanks

jt

 

[Garion]

Sure. Linksys has a combo with a 10/100/1000 switch (1 gig port, 8 10/100 ports) and a gigabit NIC for about $155. It's a 32-bit PCI card, however, which limits it's throughput. It's certainly faster than 10/100 card and gives good bragging rights. *grin*

Product link at buy.com

- G

 

[jteef]

I dont think the switches with the gigabit uplink are going to provide any kind of performance gains since I wont have a connection that exceeds more than 10mbps. I am wanting this so the computers on the lan can talk to each other at gigabit speeds.

Right now I am thinking of getting 5 gigabit nic's and a 10/100 switch with a gigabit uplink. 3 gigabit nic's would go into the gateway computer which is going to be a dual 366 celeron running freeBSD and NAT. The other two gigabit nic's would go into my computer and my roommates computer and connect to the server via crossover cable. The 3rd gigabit nic would feed to a 10/100 switch with a gigabit uplink and the rest of the network would be 10/100. But then I'd have 3 computers that would be able to communicate with each other at the limit of their respective hard drive speeds. I am not sure if my dual 366's will be strained or not when doing the routing between the interfaces at those speeds. Any ideas?

does this sound like it will work? Its about $550 cheaper than the netgear gigabit switch.

jt

 

[SaigonK]

I am going to ask...but why do you think you need a gigabit setup?
100/mb on your home lan will be fine unless you plan to run 40-50 machines in your home, and even then 100/mb is more than enough.
A 100/mb switch will do all that you want and then some...go with that instead and save your $$$

 

[jteef]

It isn't a practical need. more like a want. Why is there a ferarri dealership in manhattan? Why do people need watercooled 3 ghz p4's to read their email and post on anandtech?

I want it because it removes a bottleneck.

jt

 

[SaigonK]

Im not saying dont buy it if you want to...im just saying there is no bottleneck to speak of......you dont have any netowrk now..and going to a 100/mb will be WAY more than you will ever use at your home.
You need to get up to the 24 port switches and 3000-500 users to start saturating a lan on 100/mb.

If you think you need or want ot have a Gigabit netowrk by all means go for it, just a bit of overkill for something you will never use.
You will need to transfer files in excess of 1-2 gig to see a big performance increase...which you probably wont be doing right?
If you intend to run an Application server then yeah...go Gigiabit..

 

[Goosemaster]

I agree.

Recently I setup a Home theatersystem/network/business pc setup in his hose. He was interested in gigabit, and he DEFINITELY could have afforded it, put we chose to forgo the waste of money...why?

1. The gigabit switches/hubs you saw only have a gigabit UPLINK port. This is great for fast servers with a great deal of clients, but it provides no improvements for the actual clients..

2. THe NICs As you probably already know, are going to be on your 33mhz pci bus, which has a maximum bandwidth of 133MB as a whole. Of course, with overhead and your other dvives, you will only achieve a maximum of about 20MBps. This is up from the 6-8MBps you get with a 10/100 network. (this is assuming you don not have a server-class motherboard)

3. Why waste the money? You will never save 100hrs of hours of tranfers. Just be patient. I transfered over 40GB over my 100BaseT network On a router/switch, and it wasn't long at all.


Save your money. Trust me, this guy had ALOT OF MONEY to spend if he needed to, and he made the right decision...the smart decison.

I suggest you do the same.

Get a plasma TV or something ...he got one of those, and is getting a projector soon.

 

[PlatinumGold]

Right now I am thinking of getting 5 gigabit nic's and a 10/100 switch with a gigabit uplink.

what's the point of getting gigabit nics w/a 10/100 switch??

i personally like to limit the nics. i chose very often to use 10mb nics in order to limit the amount of traffic any single pc can flood the network with.

 

[jteef]

the point of 5 nic's would be that using 4 of them, 3 computers would all be connected at 1000mbps. The 3rd nic in the server machine doesn't need to be gigabit i guess.

I made a stupid picture of it at

Network Plan

G# are gigabit cards
S1 is a 10/100 switch with a gigabit uplink

Intel Pro/1000 T cards are $71.99 each at google gear. 71.99*5 + netgear 8 port switch w/ gigabit uplink $160 = $510 total.

or if i swap one of the gigabit nic's out for a regular 10/100 card and just get a 10/100 switch brings it down to 71.99*4 + $25+ $65 = $380

I like option 2 its about 2.5 times what an equivalent 10/100 network would cost.

I can use Free BSD to limit the bandwidth if i ever wanted to. Loud abusive language is easier though.
It is going to stress my PCI bus, especially considering the scsi hard drives I am going to have. With an X15, you figure about average throughput of 35-45mb/s. doubling for the nic is 70-90megs / second on the pci bus.

thats 3 times faster than a 100mbps network when you use it and 2.5 times the cost. Not for everybody but I dont think thats outrageous.

jt

 

[dbx]

 

[Woodchuck2000]

That looks like three gigabit cards in a single machine to me as well as a single 10/100. Which is going to be difficult to configure, and largely pointless. What on earth are you going to do to saturate a direct gigabit linkup? (which will kill your PCI bus as well btw) unless you have SCSI raid on the machine, you're never going to be able to stream enough information to kill two 100mbit links.

I would get a decent, managed 100mbit switch with a gigabit uplink, and then plug all the client machines straight into that. You would effectively get a dedicated 100mbit link to each PC from the server.

And that's if you insist on having gigabit.

I'm still managing on 10mbit at home at the moment, and that's fine for everything except large file transfers. (Going to 100mbit switched in a couple of weeks )

 

[ttn1]

I see no reason why your setup will not work. The design appears reasonable, but I'm not sure about the real performance gains.
I say, if your serious, try it out and let us know how it performs.

Although, as others have said, you have the possibility of saturating the PCI bus on the server machine.
I doubt you would ever truly saturate the bus though, because you would have to have hardrives capable of 50-60 MB/sec
sustained transfer rates. I've only seen that with SCSI RAID arrays.

I regularly transfer 3-4 GB files between machines on my home network. With 100mbps switches and nics, I get about
8-9 MB/sec sustained transfers. That is if I am doing very little on the machines other than the transfer.

I would say IDE harddrives are fast becoming the bottlenecks on home networks. Maybe you should move to 15000 rpm SCSI
drives first and then get the gigabit ethernet.

 

[Tallgeese]

Nobody has brought up the fact that getting Gigabit over Copper to work is DAMN HARD at this point if the cabling ventures outside a rack.

Expect that your cabling will have to be darn near PERFECT AND FLAWLESS for it to work optimally.
Which means a professional installation, which means big $$$$, which (no offense) doesn't sound like the case here.
I'll leave more detailed explanation on this to either spidey or ScottMac.

I will say this: If the Gig is not working optimally, then there is absolutely NO reason (short of bandwidth lust) to go with Gigabit for your setup.

Here's an idea:

High-quality 10/100 Switch (could have Gig uplink even) - This needs to support FEC or the open standard for adapter teaming. Make sure your port count is high enough.
3 dual-port Intel cards - These and the switch are configured for teaming, which can offer load balancing, redundancy, etc.
2 Cat 5 runs for each machine

BTW: What OS are the two clients using?

Just MHO...no skin off me if you wanna pay through the nose for very little benefit.

g/l

 

[jteef]

So nobody thinks this is a good idea?

A load balanced network would give me just about a 0% improvement in performance since the maximum number of simultaneous users is going to be 2-3. I am looking for an improvement in peak speed, not average. Like most home networks, on average, I predict the network traffic to be on the order of <100kbps.

Basically I look at it as a $250-300 option. I am not committed yet, but that isn't an exorbitant price IMO.

So are the standard belkin 10/100/1000 Cat5e cables not going to be sufficient in 15-25ft lengths? If they aren't sufficient, will the network auto switch to 100mbps speeds or will it just not run at the full 1000mbps rate? Belkin does sell cat6 ethernet cables that are more expensive, but I dont see a crossover cable listed. I've made a bunch of 10/100 network cables, but the terminations are usually a bit out of spec. I dont know if they'd be suitable for gigabit or not. I don't know much about the category 5e standard, but if the cards claim to work with it, I would expect them to work.

Thanks

jt

 

[Tallgeese]

now I'm really confused...

Theoretical of a dual-port NIC running full-duplex to another dual-port full NIC = 400 Mbps (yeah yeah I know no one ever sees actual throughput anywhere close to that)
Even at actual throughput rates, I have a hard time believing that your equipment could sustain and SATURATE such a link in a TINY home environment.
Your last post indicates that you expect your average traffic to be comparatively low. Which makes me wonder...why bother?

What is it you plan to do that you're looking for a higher peak speed? Is it gonna be worth $250-300?

I don't know what you plan to push around, but I've been shoving about 100GB a day up and down in a 100 full home environment and it's taking about 30-40 minutes for every 50 GB (varies based on total number of files per push).

Still, despite all that does it really matter?
cr@p man, if the money doesn't matter to ya, then try it.
not worth discussing/debating/whatever in my opinion.

For all those tuning in, however, KIM that I'm saying "go ahead' purely from an experimental, "who-cares-if-our-economy-is-in-the-crapper," unadulterated hardware lust kinda way. Wouldn't recommend this on a production network, UNLESS you could demonstrate that you actually WOULD remove a bottleneck that impeded actual, money-making work.

NOTE: Sorry for the typo above in bold. So don't PM thinking I've found a way to change the laws of physics, or that I can cr@p golden eggs

 

[Goosemaster]

It's not so much that I'm adament on prooving you wrong, as is the fact that I have already gone througth the decision you are dealing with.

For the last time, you are referring to a switch in which only the uplink port is gigabit. This is pointless for improving clienet-to-client communications, and the added cost is not worth the reduction in overhead that you speak of.

then again, why are you also talking about gigabit-type products in conjunction with this switch. THe specs even state on the page that all the ports are 10/100 except for the switch port(uplink jack). I would suggest you get a regular switch and cat 5e cabling for speed.

For reference, a REAL 10/100/1000 switch that serves gigabit to all clients is VERY expensive. expect to pay upwards of $1000 for what you want.

Negear Gigabit switch In addition, a professioanl needs to install the copperor fiber because for the copper to reach those speeds, everything must be DONE PERFECTLY, and no offense, but your handy work will not cut it. I doubt anyone here who CAN wire it properly has some Certificate or experience. I myself am not fit to preform it even though I think I can.

Please forgive me for sounding so jaded, but I have dealt with this, and was financially allowed to setup a gigabit network. In the end, I found that the price/preformance ratio is not ready for primetime(Persoanl use) just yet.

 

[spidey07]

On the other hand Gigabit ethernet is a godsend for good sized networks (200-10,000 nodes). All servers are placed on gigabit links and all inter-switch links are 1, 2 or even 8 gigabits.

If you want to play with gigabit I suggest getting two 1000BaseSX cards from intel and a 50/125 fiber patch cable. Or the linksys deal Garion posted.

Good luck. But judging from your diagram what have done is INCREASE bottlenecks and contention.

 

[Goosemaster]

Then again


If you have the money, make sure you get Scsi for ALL of your machines, and a 64bit SCSI RAID card for your server.

Might as well get the best and not live in denial

 

[Garion]

One thing to add - If you're going to be connecting a bunch of machines to your server with gigabit directly, your server will now turn into a router. When you start banging on a Celeron box with multiple gigabit interfaces, you're not going to see such great performance, ESPECIALLY when something else is hitting it at gigabit speeds. Adding the "server-as-a-router" will likely actually decrease your throughput between nodes considerably and will add some latency to each connection that's not destined for the server.

If you have $70 to burn, go out and pick up a couple of the NIC's and try it on two machines. Chances are you MIGHT get 200Mb/s out of it. Figure out if it's worth it, then buy more if it is.

IMHO, you'd get better performance by adding another hard drive and using RAID0 or a couple of new processors for your server.

G

 

[Tallgeese]

<< But judging from your diagram what have done is INCREASE bottlenecks and contention. >>

Definitely gonna be contention for that "server's" processor. All those interrupts.... :Q

 

[jteef]

Goosemaster, I realize the stuff on the 10/100 switch isn't going to see an improvement. Only transfers between my computer, my roommates computer and the "server/router" would be in a position to see an improvement. I believe the "route" command will allow the interfaces to talk to each other in a sane manner. The other computers on the network aren't going to be much...just a pc based LMS/audio test station, work laptops, etc. I did accept in Option 2 above that a regular 10/100 nic going to a cheaper 10/100 switch would probably be a good thing to do.

Is there a way to calculate just how much the server will be taxed? I think the quickest processors I can still use SMP in my bp6 are celeron 533's..<$30 on pricewatch. Is that going to be enough? I am worried about performance transferring from my computer through the router/server to my roommates. I have a monster workstation/server on my personal roadmap of projects, but the projects between now and then are numerous and expensive.

Spidey07, Can you give me a link to one of those cards and cables? I looked through Intel's site and nothing stuck out as being the right thing. Keep in mind that it will be a while before all the associated computers will have PCI-X slots. What does "50/125" mean in the realm of fiber optics?

About cable performance, All the companies advertising cards are promoting "use them in your existing networks!" Is that just marketing BS? Networks installed 2-3 years ago probably didn't even consider gigabit speeds on their cables at the time.



<< << But judging from your diagram what have done is INCREASE bottlenecks and contention. >> >>



What do you mean by this? I doubt i'd get lower speeds than a 10/100 setup.

Thanks

jt

 

[Goosemaster]

What we are refering to is the actual Harddrives on the server. I'm sorry, but IDE won't cut it at all in such a situation.

You stand to pay too much for something that is meant for other more expensive uses..let me explain.

1. It seems like you are going with the 10/100 switch so good for you.

2.SInce I am not sure if it is the one with gigabit uplink....

IF you want to have about 4 pcs taxing the server simultaneously, that comes to about 40MBps in 100Mbps or 80MBps in Duplex mode.
You will need nothing less than a Seagate 36Lp on its own or another fast SCSI Drive in RAID to provide the speed you desire.

Imagine if you had gigabit on all the pcs. That would be about 20MB/s from each pc(the pci bus won't allow you to get more speed out of it.) which equates to about the same anyways(waste of money) .
If you had workstations with 66mhz 64 bit slots, then you would would be requesting about 400MBps from the server.

DO you see where I am going with this...

If you want to get GIGABIT ONLY FOR THE UPLINK, make sure the server is fast at tranfering data. This is of course taking into account that you use your network heavily..if you don';t, scrap it.