Direct gigabit crossover connection ?

[Valhalla1]

I noticed a link on slashdot to an Addtron gigabit (copper) pci card for $35... I am interested in possibly getting 2 of these, one for my main workstation and one for my file server, and run a direct crossover connection between them. My question is, is it worth it? or is that too much bandwidth for the I/O capabilities of the machines?

the XP workstation is a 1.2tbird/512/10k rpm SCSI... but the file server is a lonely little p2-350/256/5400 rpm 80 gig

I am wondering if the I/O speed of the 5400 rpm maxtor and slow processor speed will make it so that it can't even fill up a direct 100mbit link let a lone a 1 gigabit?

 

[ScottMac]

Well, the crossover config is available at www.cisco.com, search on gigabit....it's NOT the same as for 10BASE-T or 100BASE-TX, and it's a lot more sensitive to proper configuration and component quality.

If the server can't fill a 100 meg connection, it certainly won't be any better for a gig. Even the fast machine will still have a bottleneck at the PCI bus (32 bit x bus speed, theoretical max, without considering all the other stuff going through the PCI).

Good Luck.


Scott

 

[Tallgeese]

<< Ian Addtron gigabit (copper) pci card for $35 >>

Wowsers! Cheap for Gig

<< My question is, is it worth it? or is that too much bandwidth for the I/O capabilities of the machines? the XP workstation is a 1.2tbird/512/10k rpm SCSI... but the file server is a lonely little p2-350/256/5400 rpm 80 gig. I am wondering if the I/O speed of the 5400 rpm maxtor and slow processor speed will make it so that it can't even fill up a direct 100mbit link let a lone a 1 gigabit? >>

You answered your own questions. Your server will never be able to keep that card filled, especially if it's a 64-bit PCI. For $35 per? I can't decide how important $70 is to you. But you won't see an improvement on the scale you might be expecting, if you even see any improvement at all.

 

[Tallgeese]

[OT]scott definitely posting more, and faster too, now that spidey is on vacation in the Caribbean![/OT]

 

[Valhalla1]

i usually do answer my own question but i post anyway looking for confirmation
i am not sure if my linux box fills the 100mb link, what kind of speeds SHOULD one get when ftp'ing files over a 100mbit ethernet link ?

 

[CBuxton]

ScottMac-

What's your insight into that pinout on Cisco's site for the gigabit and gigabit x-over cables? Wouldn't "regular" gigabit use the standard tia568b pinouts? wo,o,wg,b,wb,g,wbr,br....if you follow Cisco's site you get: wo,o,wg,g,b,wb,wbr,br...is this the gigabit standard, and if so, what about all the Cat6 and Cat5e jacks and patch panels? Wouldn't they be incompatible? Thanks.

 

[ScottMac]

My understanding is that the switches ports are rigged for the proper transmit/receive ordering.

The 568A standard (and the B revision) do not apply only to networking. Structured cabling can be used in applications beyond data networking, such as, phones, serial, 520-style terminal connections, 3270-style terminal connections, baseband video, broadband video limited use), alarm circuits, camera circuits, etc.

The manufacturers know and follow the 568 cabling standards, and configure their hardware/firmware/software to use the pair or pairs most suited for the application. This is the prime reason that it's important to use the correct pair ordering when terminating the cable, especially for high-speed data networking.

In the case of Gig Ethernet, all four pair are used. My recollection is that all four pair are both transmit and receive in both directions, but each pair origin has a specific destination on the other side of the connection...something goofy like that. I'll look it up and re-post with the specifics.

In my dealings with Gig-over-copper, I've never had to use a Gig crossover cable....it's not something we'd use in a business environment (all connections going to a switch). In the setup for Networld + Interop "eNet" (the Exhibitor Network), Copper Gig switches from Extreme were connected / cascaded to other Extreme copper Gig switches with straight-through jumpers....leading me to believe that (at least Extreme) switches can configure themselves or be configured for port-to-port without the need for a crossover jumper.

NIC-to-NIC seems to be a different story and perhaps Cisco switches (and/or others) do not have the crossoverless port-to-port ability. I'll check it out and re-post when I know.....maybe L3guy knows off the top of his head...I'll discuss it with him tomorrow (or maybe he'll jump in and post what he knows......).

BTW: in a structured cabling environment, all cable in the wall is configured pin-for-pin (straight through). Any reconfiguration or adaptation is done external to the standing cable plant (with jumpers or adapters).

EDIT: I found this link from to the UNH Interoperability lab. It has some excellent information on the nuts and bolts of Gig E over Copper (1000BASE-TX).

1000BASE-T White papers from the UNH Interoperability Lab.

FWIW

Scott

 

[Valhalla1]

lotsa great info scottmac, the elite title is well deserved

... can anyone tell me what kinda speeds you get when you ftp files over a 100mbit link? I would like to see if I have a bottleneck somewhere that I dont know about..


here's what I get, ftp'ing a file from the redhat box to XP

DU Meter showed 6.3-7.0Mbytes/sec, it stayed around 7MB (56mbit) so I am only filling a little over half of the 100mbit link as it is now

now, I was running ECCP distributed computing client on the linux box while transferring, so the cpu may not have been able to handle the packets as fast as it could.. tried it again with eccp stopped, and it was actually a LOT more random, fluxuating from under 10mbit up to 65 peak (only for a couple seconds)... this is no good, i'm afraid something is wrong.

i normally make my own cables but the linksys 10/100 switch came with some pro-made ethernet cables so I used those for the file server and XP box (dad's switch link gets a home-made ) so i dont think its the cables or switch.. maybe linux isn't liking the SMC ez-car 10/100 nic.. although running 'ifconfig' shows 0 errors/collisions/dropped packets, same on XP





*EDIT* okay, I tried using iperf, a CLI network benchmarking tool, and it tells me this:

[1972] local 192.168.0.1 port 3072 connected with 192.168.0.90 port 5001
[ ID] Interval Transfer Bandwidth
[1972] 0.0-10.0 sec 99.9 MBytes 79.9 Mbits/sec


which is much better than what I was showing for FTP.. so who knows

 

[Nutz]

What I want to know is if the GigE as we know it running on CAT-5e with all 4 pairs is 1000BaseTx, or 1000BaseT4. From what I've read its the latter and the TX variant will run on CAT-6 with only 2 pairs a la 100BaseT. So if what I've read is in fact correct, wouldn't the crossovers work/wired the same as they are for crossovers in 100BaseT?

 

[ScottMac]

I am not aware of a 1000BASE-T4 spec, only 100BASE-T4. The only other copper Gig spec I've heard of is 1000BASE-CX, across Shielded Twisted Pair (NOT screened UTP).

Since Gig over copper came out, it has operated on all four pair. When Gig over copper hit the street, Cat5e & Cat6 were scarce, at best (Cat6 was still pending ratification, and 5e was pretty much brand new).

I'd bet that things will stay as they are as far as how Gigabit Ethernet is propagated across UTP. But, I'm just guessing here....

As far as 100BASE-T LAN throughput, I believe you're probably in the ballpark, given the information you've provided.

I'll try to figure out how I can pass along a copy of "Bricks" to y'all. It's a program written by Digital Equipment Corp (DEC) many years ago to demonstrate ATM throughput. It operates from RAM through the NIC to the other machine for display...so the disk I/O is removed from the picture.

Bricks basically shows a pile of bricks, each brick represents a certain number of bits (from 100,000 to 10,000,000, user configurable). The transmitting machine removes a brick, sends it down the pipe, where it is received and put either on another pile, or to build a bridge (it's a show demo program). Multiple transmitters can send to a single receiver, each transmitter sends a different color brick. With multiple transmitters, the receiver display shows very nicely the relative throughput of each transmitter.

As soon as I can find a public place to put it, I'll upload it and post the link with instructions.

FWIW

Scott

 

[Tallgeese]

<< lotsa great info scottmac, the elite title is well deserved >>

Another entry for "Understatement of the Year" right there

 

[Valhalla1]

scottmac - how large is the file, I can host it for you if its not too big

 

[ScottMac]

It's like ~165K ... probably emailable. The other part of the delay is that I have to write up instructions for use....it's not hard, but there are some "features" that must be set up just right for it to work. I can probably have it ready in a day or so, then I'll probably just put it in Yahoo briefcase or something....or I'll send it to you if you want to put it up.

BTW: Another throughput/performance utility that works well is Qcheck, from the NetIQ website. It's free, and the baby brother of "Chariot," a seven-layer test suite that we used in the lab. Check it out.

Thanks for the offer, I'll letya know, probably tomorrow or so.

FWIW

Scott