Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I have two Netgear GS748TS 48 port gigabit stackable switches, which when stacked via two HDMI cables allow hosts on one of the switches to talk to hosts on the other.

Netgear claim that the HDMI trunk allows for 10Gbps of bandwidth, but when I test using 4 computers and iperf, it appears that only 1Gbps of bandwidth is available for communication between the switches.

Is additional configuration required? Am I missing some basic networking concepts here?

I'd really like to keep all of the client ports for, well... clients, rather than sacrificing 16 or more of them for a trunk.

Any help is appreciated, many thanks.

EDIT: I'll re-test with higher quality HDMI cables and report back!

share|improve this question
    
so does this mean that the gs748ts is truly stacked as in a single forwarding plane, sorry don't have time to read the specs on the device but if it does then thats cool. –  tony roth Jul 18 '12 at 18:37
    
Judging by its performance and price point I dont think so –  ilumos Jul 18 '12 at 18:43

5 Answers 5

You need to ensure you have a HDMI v1.3 or 1.4 compliant cable to achieve this, even then you'll never see more than 8.16Gbps due to overhead.

I've never seen HDMI being used this way, most sysadmins just buy switches with 10Gig ethernet ports to deal with the...erm, 10Gig ethernet. I like the ideal of the HDMI but it sounds a bit fragile a solution (as is the case here) and it seems a bit 'prosumer' to me. Imagine it's cheap though.

share|improve this answer
1  
Netgear and Dell, and likely others, have started using HDMI cables between switches. HDMI and SGMII are nearly identical at an electrical level, so it's pretty easy to do. –  Chris S Jul 18 '12 at 17:53
1  
@ChrisS I can totally see why this will be used by small/medium sized business but I can't see it taking off in enterprises, can you? yet I'd happily use thunderbolt due to the inherent high-ended'ness of even the cheapest TB cables. –  Chopper3 Jul 18 '12 at 18:03
    
The HDMI cables I bought claim to be HDMI 1.4, but they weren't expensive... They're also very short at 1m in length, I'd expect cheap poor quality cables to perform badly at long distances but not at 1m. –  ilumos Jul 18 '12 at 18:26
    
Try others..... –  Chopper3 Jul 18 '12 at 18:47
    
@Chopper3 I agree, I don't think this will take off in the higher end equipment, but for cost conscious consumers it'll probably be a hit. Agreed, try other cables.. especially if they're from the 'far East' as they're probably not up to 1.4 spec. –  Chris S Jul 18 '12 at 19:21

How are you testing exactly? It sounds like its your testing method at fault, not the switches or cable. The fact its rounding off to a neat 1gbps would dismiss the claims the cable quality is at fault.

I guess you are either testing with either,

  1. 1 server and 3 clients
  2. 2 servers and 2 clients

The former will always be limited to the connectivity of the server, so if its a non bonded (LACP) connection, you'll only ever see a 1gbps cumulative result.

The latter, if both tests were run simultaneously, should see a total of 2gbps throughput across both iperf tests. So if you're not getting this result, but are using this method - then it sounds like a config issue.

Surprisingly, the netgear forums for the prosafe kit are pretty useful, they have a few mods on there who are higher level netgear techs; so its definitely worth pursuing a solution there too.

share|improve this answer
    
Testing using the latter. 4 computers total, 2 as iperf servers, 2 as iperf clients, with 2 computers on each of the switches. Each server/client pair can achieve gigabit speeds when only one pair is testing, but setting both going at once results in one of the pair tanking down to about 15mbps. –  ilumos Jul 18 '12 at 20:49
    
Okay. I just had a thought. Have you got auto-dos turned on? I've seen it cause problems with a number of protocols and iperf in particular. –  Sonassi Jul 18 '12 at 21:45
    
I've had a look through the manual and found no DoS prevention settings –  ilumos Jul 19 '12 at 6:31
    
The fact its rounding off to a neat 1gbps would dismiss the claims the cable quality is at fault. - or that the switches auto-negotiate at their highest speed and then fail the speed down until they find one that works. –  RobM Jul 19 '12 at 8:07

the stacking ports are full duplex so a single cable completes the ring.

A hmdi plug is probably stronger than a LC fibre plug, but as the HDMI loop is in the rack (I doubt you could run a 10m HDMI and have a working stack) strength/latching capacity it a bit irrelevant.

I would say, having bought stacking kits in the past for several hundred pounds a £3 HDMI lead from CPC suites me fine!

As for dodgy test topology, I don't see how you can test a 10GB link without 10GB infrastructure, to find true speed you would surely need to have a couple of servers with 10 concatenated 1GBs connections, I suspect as above the test topology used showed results for a single port, as I fail to see how it could scope the backplane bandwidth.

share|improve this answer

Have you tried teaming ports between the two ends? You will definitely only get 1 gigabit if you're using a single port. If you have computers with multiple gigabit network cards, try setting them up in a team, then set up the switch with the ports in an LACP trunk, and then run iperf between them.

I see you said that speed dropped when two iperf sessions ran at the same time. This test will eliminate the possibility of the cable being faulty.

share|improve this answer

a HDMI lead is just a neat version of a LVDS link, you could use CX4, U320 VHCDI or even DVI, first, you say 2 HDMI leads, I assume you have not used both! you stack a switch with a single cable.

HDMI is probably tougher than LC fibre so really, what's the issue, I have 2 GS748TS' stack and can easy move 5gbs across them (I have a 4gbs LAG from a quad card in my file server feeding one switch and the render farm on the other can pull over 380MB/s from that server...

I think you have a dodgy testing topology IMHO

share|improve this answer
    
Most modern communications cables use LVDS, sure. The stack needs to be in a ring, so both connections will be used. HDMI is "tougher" than fiber? On what measure? How is that even relevant? "Dodgy testing topology"? How exactly? –  Chris S Jul 18 at 14:00

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.