Slashdot Mirror
Gigabit Networking for the Home?
The Clockwork Troll asks: "I've had a whole-house audio/video distribution project on the back-burner for a while now. As gigabit networking hardware prices come down to earth, I'm tempted to jump on the 1000BaseTX bandwagon. As far as I can tell though, the current crop of consumer-priced hardware/software doesn't address a couple key issues, namely: fragmenting jumbo frames for the benefit of legacy clients - this is critical as some of the devices on my network will not tolerate the 9000+ byte Ethernet frames which are needed to get the most out of gigabit; and OS support - do Linux and Windows require much tweaking to take advantage of gigabit? Will most drivers automatically optimize themselves? A Google search didn't reveal too much consensus, especially on hardware choices. What switches and software configurations have Slashdot readers been using for home gigabit networks, in particular mixed ones (100/1000BaseTX?"
That's gigabyte. Bit prefixes go by 1000 (10^2), byte prefixes go by 2^10.
Of course, if your needs are more extensive you may need something more...
Al Qaeda has ninjas!
Now I know this is /. but before everyone says "you don't need gigabit!" and "bah, who needs that kind of speed" gigabit ethernet is genuinely useful. Even copying 500mb files can take intolerably long when you want it done 4 minutes ago. If the poster wanted a bunch of nonsense about why he shouldn't do it and why its a dumb idea, he could have gone to Circuit city (they don't sell gigabit so they would try to sell him 10/100). Instead he asked us for an informed option and information on the matter.
Even those who arrange and design shrubberies are under considerable economic stress at this period in history.
If you regularly copy videos for editing, gigabit is great. Many homes do this. We do this. With the price so resonable, I can't see why anyone who does video work wouldn't get gigabit.
TW
I was fortunate enough to buy 2 TI-Chipped Firewire cards ($20 total) and use them to network my main WS to my Server w/ 5-foot cable. You can save a lot of money going this route if you can. MM
But you don't have to be able to pump the full 1000 Mbps to take advantage of gigabit ethernet. As long as you can pump more than 100 Mbps, then gigabit will give you a speed improvement over 100 Mbps ethernet.
(Or, a good location for the ceiling is "anywhere above your head").
"The same goes for switches. You'll be doing good to get 400 mbps out of a cheap gig switch."
40MB is a hell of a lot better than 10MB. I don't know why everyone keeps saying he won't be able to saturate the line. He doesn't need to max it out in order to enjoy the benefits over 100Mb ethernet. Who knows what kind data we will be dealing with in 5 years? Seems like going 1000 is a smart investment.
I had no idea Gb Ethernet switches had dropped so much in price. If I was buying a new switch today I'd definitely be buying one of those $100 Linksys switches. Considering the cost is so cheap why even bother with 100MB if you think you'll be using bandwidth hungry apps?
If you wanna get rich, you know that payback is a bitch
Being integrated with the motherboard doesn't make a performance difference on any board I've ever seen. It still goes over the PCI bus, it's just not using a slot. Creating a separate bus just for the ethernet port would be too expensive.
I'd be interested to know where you came up with that. Some switches may have an underpowered backplane that limits your aggregate bandwidth (such that you can't pump a full 1Gbps on all ports simultaneously) but it shouldn't prevent you pushing 1Gbps between two ports when all else is idle. If it's advertised as a gigabit switch but is only capable of 400 Mbps, wouldn't the manufacturer be open to claims of false advertising?
Uh, have you ever tried copying 3 hours worth of HD video from your capture system to your main workstation over 100 Mb. After 15 minutes you will be begging for Gb.
Even though I would probably only get double the speed (disk bottlenecks, one is a slow system) I still am thinking it might be worth it.
Q.
Most ATA HDD's can transfer around 40-60 MBps. You can easily saturate a 100BaseT network with bargain basement machines.
I beg to differ. The numbers you quote there are empty benchmarks of an ATA drive alone within an OS and a benchmark tool, or some OS-less independent method devised by manufacturers across and IDE bus. I said the drives couldn't handle it.. any sustained transfer at that rate, even if the drive would support a streamed write for a sustained period, being fed at "good" gigabit speeds of ~800Mb/s, would surely melt the drives. But in practical terms -- (I should not have said just the drives, earlier) -- the pure drive metrics are useless for this discussion. You must take into consideration that all of the following will destroy every good number you might have had: the data (or file) transfer method, the capabilities of the OS itself, and how it's tuned, the application in use, and how it handles checking, transmits, and writes, any number of ethernet based faults, retransmits, etc, IP fragmentation, packet reconstruction, TCP window size and frag size tuning (or lack thereof), the position of the moon at the start of transfer..
IMO gigabit in house is a waste. Take the money you would have spent on a switch and NICs and buy some good champagne and cigars.
I dont get it. People here are bitching that the best throughput they see on gigabit ethernet is 400Mbps. Thats 4x the speed of regular 100Mbps ethernet. 4x still seems like a hell of an improvement, especially when you consider gigabit switches can be had for $100-150. I'd take a 4x faster HDD, processor, memory, etc anyday! Why snub your noses at at 4x network speed increase?
'I don't think anyone needs more than what I've used'. Please, put a bullet in your head.
Because you'll find that you can't write to a filesystem on a single disk much faster than 100mbit anyway. Gigabit is significantly faster than the I/O that a single drive can provide.
Government of the people, by corporate executives, for corporate profits.
They can read at that speed. They can't write at that speed. You will need large memory buffers (similar to the size of the files) on either end of the network to handle the slowdown when waiting for the disk, or a stripe across several disk spindles.
Then of course for smaller files there's the seek times, you don't get anything like the maximum theoretical throughput from the drive. As to waiting for 1/4 of the time, it depends whether it's 0.01s or 60s.
Government of the people, by corporate executives, for corporate profits.
HD is a wasteland right now. Some of the networks are in HD some of the time, if network sitcoms and a few sporting events is your idea of watching TV. There's HBO and Showtime, if you get either one, and then there's a PBS and a Discovery HD which are almost just a loop. Beyond that and the re-hashed crap on HDNet there really isn't anything terribly compelling in HD.
...what kind of bandwidth streaming DVD content actually needs. It's 6-9Mbps which means you could probably serve 6-8 "terminals" simultaneously with a freaking 100Mbit hub. With a 100Mbit switch, it's a no-brainer.
He's also clearly clueless about consumer hard drive transfer speeds. Unless the OP has a nice RAID setup on his server (which is doubtful if he's un-savvy enough to come to Slashdot for networking advice) there is no way that this theoretical 120GB workstation backup will be appreciably faster. I say this as someone who actually has Gig the Fast Ethernet running on a home network with a server that uses ATA drives in both RAID and non-RAID configurations, not as someone who plugs numbers into a calculator and makes ridiculous assumptions.
cybermace5, stop trying to spend someone else's money based on your incomplete understanding of both the requirements and the technology.
Hint, that's the maximum theoretical read performance. Hard disks read significantly faster than they write.
/dev/zero.
Test it on your system:
Reading:
dd of=/dev/null if=/tmp/file bs=64k count=131072
Writing:
dd if=/dev/zero of=/tmp/file bs=64k count=131072
You should try it with different count values to see how your filesystem buffer affects the speed. Every file you read has to be written somewhere (unless streaming video for instance) and when you have very large files (e.g. 4Gb) your filesystem buffer will be flushed through unless you have configured a 4Gb buffer of course. To take any sort of advantage of gigabit, you need large enough buffers to make sure you aren't being limited by the write speed of the receiving drive.
I predict that you won't get anything like the 32Mb/s quoted, never mind 58Mb/s once you're running at the disk speed rather than the buffer speed. Even with the ideal condition of dd'ing from
Government of the people, by corporate executives, for corporate profits.
Geez what do you want. There is more programming available in HD now, than there was OTA programming 20 years ago.