10 Gigabit Networking

[nicklmg]

Amazon: http://geni.us/Ufw
NCIX: http://bit.ly/1Qjmxju

 

I decided it was time for a networking upgrade here at the office, and I would NOT settle for second best. So, full 10 gigabit it is!

 

[rattacko123]

first!

[Ridska]

 Dammn son ! 10 giga net ?!

linus , u be balllin'.

[tt2468]

How that pfsense box doin? Did you get the sfp card to work?

[PortentousLad]

wanna bet it's windows?

 

 

Cool stuff Linus! keep 'em cumin'  (hihi... "cumin'" originally a typo, but I'm leaving it in cos it's funny)

[Akira4950]

Think I know what the problem is: crosstalk. LinusTechTips has a high density in terms of its network load at any given time. Crosstalk is more likely to occur when you have a lot of cables bundled together. Although he is using cat6a cables, he probably bought the cheapest available. Cheap cables usually mean lower quality shielding. Also Linus is probably at the maximum when where come to transferring 10 gigabit on cat6a (100m). Although I don't know the drop off rate for cat 6a. I do know that when cats for crosstalk is high cat 6 cables the effective length for transferring 10 gigabit drops from 55 meters to about 30 meters. About 40% reduction.- might want to try to use cat 7 or cat 7a cables. I think cat 8 has been finalized but I doubt you'll be able to find anything at a reasonable price. And again if crosstalk is the issue the effective length for transferring 10 gigabit with cat 6a cables, using the 40% reduction, should be about 60 meters or 180 feet.

Edit I was thinking that some evidence if the 40% reduction is showing up in data rates. If he can only get 600MBps he might be able to test my theory by running one cat 7a ( terminated withGG45 collectors) to the computer that's having the worst connection. the cable would need to run with the cat6a cables to rule out crosstalk.

[tbmatuka]

Why are there no links for all the components?

 

2-port Intel adapter they used: X540-T2

1-port Intel adapter: X540-T1

12-port Netgear managed switch they used: XS712T

8-port Netgear (mostly) unmanaged switch: XS708E

[Snadzies]

There is nothing more hair-pullingly-frustrating than networking issues.

Makes you just want to scream "WHY WON'T YOU WORK?!" and punch something.

[Geralt]

Gotta go fast.

 

Sanic the Firefox: 10Gbit Edition

[dajinn]

Nice video, it's cool to see LTT going in this direction. I would prefer to see more "enterprise-grade" based videos instead of all the gamer stuff.

 

That being said, the video is a little disingenuous.

 

10 gigabit networking in the home is still not cost effective. Unless your pockets are lined by sponsors or you have a very meaty salary, 10 gigabit RJ45 ethernet networking is a hard sell. The 8 port switch goes for around 760~, and the 12 port even more at a whopping 1200 bucks. Neither or which have really saturated the second hand markets yet.

 

Additionally the network cards themselves are pretty costly. I'm no stranger to the best offer system on ebay and recognize it's how I got a lot of my sweet deals. There's no telling how low of an offer the north american sellers of the X540-T2 cards will go but I wouldn't bet anything under 250. Maybe 200 if they've having a slow month. That's still a grand to outfit just 5 machines.

[Downloadski]

Some nice tips with the buffers, got to try that. I have intel 520-da2 cards with DA cables into a d-link 28 port switch with 4xSFP+ slots.

Have better send than receive speed, so perhaps use some settings show in the video.

[Jaip]

I would recommend setting the Jumbo frames equal at every end. Meaning at the switch, client and server side. The MTU (Maximum Transmission Unit) needs to be the same:
When the clients (NIC's) are set lower (x), the pc needs to chop (fragment) the data into the size it is set to (x), while the switch can handle more (y).
Also a mismatch between those two could cause errors (datapacket drops etc) which results in worse connection (server sending larger size's, which the switch can handle, but the NIC on the client is set lower => fragmentation of the packet.. loss of speed)

The inconsistency I can't explain immediately.. Does it occur with the same file?

[OhYou_]

Dammit, and I just switched to 1gbit at home.

[rmac52]

The colors seam a little washed out an soft.  Not really complaining just pointing out an observation.

[pk-man]

the lighting seems weird. It looks like it's shot in a front of green screen

[ggcs.tk]

Any prices on that kit?

[unijab]

I like videos like this and hope to see more.

[Cyran]

Could you have used Thunderbolt drives, and just add adapters to your machines?

[leadeater]

Think I know what the problem is: crosstalk. LinusTechTips has a high density in terms of its network load at any given time. Crosstalk is more likely to occur when you have a lot of cables bundled together. Although he is using cat6a cables, he probably bought the cheapest available. Cheap cables usually mean lower quality shielding. Also Linus is probably at the maximum when where come to transferring 10 gigabit on cat6a (100m). Although I don't know the drop off rate for cat 6a. I do know that when cats for crosstalk is high cat 6 cables the effective length for transferring 10 gigabit drops from 55 meters to about 30 meters. About 40% reduction.- might want to try to use cat 7 or cat 7a cables. I think cat 8 has been finalized but I doubt you'll be able to find anything at a reasonable price. And again if crosstalk is the issue the effective length for transferring 10 gigabit with cat 6a cables, using the 40% reduction, should be about 60 meters or 180 feet.

Edit I was thinking that some evidence if the 40% reduction is showing up in data rates. If he can only get 600MBps he might be able to test my theory by running one cat 7a ( terminated withGG45 collectors) to the computer that's having the worst connection. the cable would need to run with the cat6a cables to rule out crosstalk.

 

Fairly sure the cabling being used is UTP (unshielded) and not S/FTP or F/UTP etc. Also the Cat standards are strict gradings, you meet them or you don't, cheap cable or not. This is similar to the myth of HDMI premium cables. Also LMG cabling and data rate density isn't that high.

 

I run 10Gb from my desktop to server at home using Intel X540-T1 directly connected, no switch, using self terminated Cat 6 cabling over a distance of about 40m-50m. I can saturate the link at full 10Gb using SMB3 and iSCSI. 

It's hard to give good networking advice by just watching a video and listening to a description of what has been tried but my advise to @LinusTech would be to use iperf to test the network link, would show if it is actually a network throughput issue or something else. 

 

Edit: Woops he's already used a similar tool, teach me to reply to a post about a video I haven't watched in ages then re-watch it after posting.

[TWB]

http://www.overclock3d.net/news/gpu_displays/hidden_gems_of_ces_asus_10gb_networking/1

 

Asus about to release affordable 10GBase network switch (300$) XG-D2008 and network cards...