Is gigabit ethernet usually hot?

[DanceswithUnix]

I finally cracked a short while ago and bought some gigabit ethernet stuff from Scan as it was on offer and I was getting frustrated with the speed of 100Mb.

I was expecting the 5 port switch to get a tad warm, but plugging the server into the switch made the fan in the server PSU start whirring faster.

When I took the side off the server and tried prodding the (admittently cheap Realtek chipset) PCI network card it was damned hot. Too hot to hold my finger on, so probably about 70 degrees.

I have added a quiet 120mm inlet fan on the front of the case, so the PSU fan is settled back down now. I still wonder if gigabit usually runs that hot though, this is the first time I have had a chance to use it.

I have a power meter on the mains socket that runs the server and network gear, and plugging/unplugging the server into the switch causes a jump of about 9W (I assume that is 4.5W each in switch and server).

[arthurleung]

No. Basically, running at very high GbE speed (i.e. 70MB/s) will be full loading the CPU (NF4 Lan) Since you are using Realtek GbE, I would think 35MB/s will max your CPU already.

If your CPU is at max, it means more heat, and more power drain, therefore your PSU crank up the fan to cool it.

Its not the GbE chip using the electricity, but the CPU.

9W is probably your extra CPU load. Switch doesn't get hot most of the time, only warm at most. I have tried pushing 100MB/s across the network (4 computers, 1 to 1 transfer) for a few hours (just to test max bandwidth) and my switch is not any warmer than idle

[Butcher]

Why would 70mb/s heavily load the CPU when a CPU can move several GB/s?

[RavenNight]

It may be todo with the effort the CPU has to do to move the stuff on a lan.

[arthurleung]

Why would 70mb/s heavily load the CPU when a CPU can move several GB/s?

Network overhead, poorly writen driver / tcp implementation, etc.
You could try it yourself. I do see my explorer pop-ing up to 100% when I'm sending across the network (And the load is directly proportional to transfer rate). Definitely not disk problem because locally it reads at 230MB/s and only take 5% CPU max.

[latrosicarius]

Why would 70mb/s heavily load the CPU when a CPU can move several GB/s?

I think he means the processor on the cheap Realtek LAN card can only move 70mbps, not the system CPU. Though, if it can't move at least 1Gbps, then it's not a Gigabit LAN card and it don't matter what kind of switch he's using... he won't get gigabit.

[arthurleung]

I think he means the processor on the cheap Realtek LAN card can only move 70mbps, not the system CPU. Though, if it can't move at least 1Gbps, then it's not a Gigabit LAN card and it don't matter what kind of switch he's using... he won't get gigabit.

I hope I didn't confuse you.
I did not state mbps but MB/s and I do mean 70MB/s (560Mbps)
And there is not a processor on the Realtek LAN card. You can find one on an Intel card but it only handle some of the network overhead but the CPU still need to process a lot of stuff for the data to go through the network.

[DanceswithUnix]

No. Basically, running at very high GbE speed (i.e. 70MB/s) will be full loading the CPU (NF4 Lan) Since you are using Realtek GbE, I would think 35MB/s will max your CPU already.

If your CPU is at max, it means more heat, and more power drain, therefore your PSU crank up the fan to cool it.

Its not the GbE chip using the electricity, but the CPU.

9W is probably your extra CPU load. Switch doesn't get hot most of the time, only warm at most. I have tried pushing 100MB/s across the network (4 computers, 1 to 1 transfer) for a few hours (just to test max bandwidth) and my switch is not any warmer than idle

That's an interesting theory, but in my case I don't think it fits for three reasons:

1/ The power goes up with the network is idle.
2/ My server uses old hand-me-down components, in this case an Athlon 1000 with basically no power management (well, that works anyway).
3/ I nearly burnt my finger on the network card.

I expect the CPU is drawing it's rated 52W _all_ the time.

I do notice that the 5 port switch is only a bit warm, despite handling 3 Gbe connections on top of 1 100Mb connection (if it was a cup of tea, I would call it warm).

Just tried a file copy, and the power usage does go up by about 15W whilst managing 33MB/sec. That is split between workstation, server and switch though, so that could be all in the workstation which is athlon64 and so does most certainly have power management. Having said that, the A64 on idle can probably run rings around the athlon 1000

[arthurleung]

That's an interesting theory, but in my case I don't think it fits for three reasons:

1/ The power goes up with the network is idle.
2/ My server uses old hand-me-down components, in this case an Athlon 1000 with basically no power management (well, that works anyway).
3/ I nearly burnt my finger on the network card.

I expect the CPU is drawing it's rated 52W _all_ the time.

I do notice that the 5 port switch is only a bit warm, despite handling 3 Gbe connections on top of 1 100Mb connection (if it was a cup of tea, I would call it warm).

Just tried a file copy, and the power usage does go up by about 15W whilst managing 33MB/sec. That is split between workstation, server and switch though, so that could be all in the workstation which is athlon64 and so does most certainly have power management. Having said that, the A64 on idle can probably run rings around the athlon 1000

A typical network card use about 3W itself. Don't underestimate the 3W. There is no heatsink on the chip so the 3W builds up quickly. Hot != using lots of power. If you put in 3J every second, but the chip does not release 3J to the surrounding then the temperature increases Simple physics.

To clarify about the power management. There are two forms of them. One is to tell the CPU to idle (Works on all CPU), and one is to actually reduce the frequence/voltage (Athlon 64, Pentium-M, Pentium 4 to a certain extent). Your CPU will not use 52W all the time. If it does, the idle temperature will be exectly the same as full-load temperature. You can check that yourself.