I've had a standard wired home network for awhile now using a Dynex (I believe D-Link rebranded) 4-port router with regular Cat5 cables and RJ-45 jacks. The LAN connection in XP on all my systems claims 100 Mbps. Is that true, and does that mean I should theoretically be able to transfer at 12.5 MB/s between all my computers?
Is a standard wired home network 100 Mbps?
- Thread starter archcommus
- Start date
- Oct 25, 1999
- 29,554
- 430
- 126
Yap, the current Entry Level devices are 100Mb/sec.
However real life speeds are the following, http://www.ezlan.net/net_speed.html
:sun:
However real life speeds are the following, http://www.ezlan.net/net_speed.html
:sun:
Home internet access is of course typically bottlenecked well below 100 Mb/s -- let's assume we all understand that with normal internet bandwidth, your home LAN speed isn't going to make much of a difference.
Home and corporate networking is to a large extent using 100 Mb/s. 12.5 MB/s is the theoretical limit not considering overhead, but in practice you're going to get closer to 10 MB/s file transfer performance at best. But, IMO, gigabit is so common and affordable, that if you care about internal network performance, then you should regard gigabit as the easily-available standard, and 100 Mb/s as yesterday's technology.
You can get a home GbE switch for around $35, and most MB's have GbE NICs built-in, so these don't cost you any extra these days. You can get good Intel add-on NICs for as little as $21.
For as little as that, you can get internal file transfers at around 30 MB/s -- this is 3X as fast as you can do with 100 Mb/s ethernet. Even if you only get 20 MB/s, this is 2X as fast as 100 Mb/s ethernet. How much performance gain would you get with SLI/Crossfire for example? You'd get less than 2X, for a lot more money. In this view, even the low-end 2X performance improvement with gigabit is a bargain, with the caveat that:
(*) This is conditional on internal file transfer performance actually mattering to you. If all you're doing is internet browsing, email, etc., at 10 Mb/s or even high-end 30 Mb/s for example, then having an internal gigabit network isn't going to get you much benefit. If you're transferring 100's of MB's regularly or using a file server regularly, then it will make a difference.
To illustrate, imagine that you're accessing data across your internal network, and waiting for the results. 10 MB/s stands for good 100 Mb/s ethernet; 30 MB/s stands for typical gigabit:
10 MB @ 10 MB/s = 1s
10 MB @ 30 MB/s = 0.3s (swamped by user time; can't really feel the difference)
100 MB @ 10 MB/s = 10s
100 MB @ 30 MB/s = 3s (still swamped by user time, but might feel the difference)
300 MB @ 10 MB/s = 30s
300 MB @ 30 MB/s = 10s (can feel the difference)
600 MB @ 10 MB/s = 60s
600 MB @ 30 MB/s = 20s (can certainly feel the difference, and you'll start to want faster than 30 MB/s now).
In other threads here, I've seen several claims that you must have high-end gear / OS's, etc. to make gigabit worthwhile. I'd be happy to discuss those claims here, but to be fruitful and not just empty differences of opinion, we'd need specifics -- which tests have shown for example that a $35 D-Link DGS-1005D is wholly inadequate, etc., that jumbo frames are critical, how much better performance can you demonstrate with for example a certain $100 GbE switch, etc.
My premise is that 2-3X 100 Mb/s performance is commonly achievable with inexpensive consumer gigabit, and assuming (*), that internal LAN performance matters to you, it's worthwhile. If you don't really care about internal LAN performance, of course none of this is relevant to you.
Home and corporate networking is to a large extent using 100 Mb/s. 12.5 MB/s is the theoretical limit not considering overhead, but in practice you're going to get closer to 10 MB/s file transfer performance at best. But, IMO, gigabit is so common and affordable, that if you care about internal network performance, then you should regard gigabit as the easily-available standard, and 100 Mb/s as yesterday's technology.
You can get a home GbE switch for around $35, and most MB's have GbE NICs built-in, so these don't cost you any extra these days. You can get good Intel add-on NICs for as little as $21.
For as little as that, you can get internal file transfers at around 30 MB/s -- this is 3X as fast as you can do with 100 Mb/s ethernet. Even if you only get 20 MB/s, this is 2X as fast as 100 Mb/s ethernet. How much performance gain would you get with SLI/Crossfire for example? You'd get less than 2X, for a lot more money. In this view, even the low-end 2X performance improvement with gigabit is a bargain, with the caveat that:
(*) This is conditional on internal file transfer performance actually mattering to you. If all you're doing is internet browsing, email, etc., at 10 Mb/s or even high-end 30 Mb/s for example, then having an internal gigabit network isn't going to get you much benefit. If you're transferring 100's of MB's regularly or using a file server regularly, then it will make a difference.
To illustrate, imagine that you're accessing data across your internal network, and waiting for the results. 10 MB/s stands for good 100 Mb/s ethernet; 30 MB/s stands for typical gigabit:
10 MB @ 10 MB/s = 1s
10 MB @ 30 MB/s = 0.3s (swamped by user time; can't really feel the difference)
100 MB @ 10 MB/s = 10s
100 MB @ 30 MB/s = 3s (still swamped by user time, but might feel the difference)
300 MB @ 10 MB/s = 30s
300 MB @ 30 MB/s = 10s (can feel the difference)
600 MB @ 10 MB/s = 60s
600 MB @ 30 MB/s = 20s (can certainly feel the difference, and you'll start to want faster than 30 MB/s now).
In other threads here, I've seen several claims that you must have high-end gear / OS's, etc. to make gigabit worthwhile. I'd be happy to discuss those claims here, but to be fruitful and not just empty differences of opinion, we'd need specifics -- which tests have shown for example that a $35 D-Link DGS-1005D is wholly inadequate, etc., that jumbo frames are critical, how much better performance can you demonstrate with for example a certain $100 GbE switch, etc.
My premise is that 2-3X 100 Mb/s performance is commonly achievable with inexpensive consumer gigabit, and assuming (*), that internal LAN performance matters to you, it's worthwhile. If you don't really care about internal LAN performance, of course none of this is relevant to you.
Here are some of my stats:
Directories processed = 15,604
Total data in bytes = 1,148,731,401,763
Elapsed time in sec. = 21,102
Files copied = 290,508
1.148 TB / 21,102s = 54.4 MB/s
This was using built-in NIC's and a consumer GbE switch without jumbo frames enabled (D-Link DGS-1008D, the 8-port sibling of the switch mentioned above).
Directories processed = 15,604
Total data in bytes = 1,148,731,401,763
Elapsed time in sec. = 21,102
Files copied = 290,508
1.148 TB / 21,102s = 54.4 MB/s
This was using built-in NIC's and a consumer GbE switch without jumbo frames enabled (D-Link DGS-1008D, the 8-port sibling of the switch mentioned above).
So I'm getting about 8-10 MB/sec, then. Okay thanks.
Seems kind of ridiculous that you can't get anywhere near 125 Mbps with standard consumer gigabit equipment, though.
Seems kind of ridiculous that you can't get anywhere near 125 Mbps with standard consumer gigabit equipment, though.
You can. I've measured in excess of 115 MB/s sustained using consumer gigabit without jumbo frames -- this is great considering that there's also TCP/IP protocol overhead to consider.
It's the HD's + file transfer protocol overhead that can't fill the pipe with data at that rate, which make typical throughput lower. This is more complex. Even there you can help things along with fast drives, RAID arrays, and good/clean file systems if you want to take the trouble and cost.
Above, I demonstrate > 5x the throughput of 100 Mb/s "fast" ethernet, for > 1 TB of data transfer -- is this really something to complain about, esp. considering how inexpensive the network hardware can be? The NICs come included with MB's, so are almost "free". In many cases, the only thing you need to add is a GbE switch, and these can be inexpensive, as I also show above. What's there to dislike?
Of course you can get greedy and want more; my basic point is that what you get for next to nothing in cost is great -- even a 2x performance improvement is really substantial if it matters to you, if you consider it relative to performance gains you can get elsewhere.
Getting greedy can also cost you more in having updating your HD performance, etc., etc., and the easy solution there is to not get greedy, and take the 3x or 2x performance improvements you can easily get with satisfaction.
I agree that that performance is good.
It can be really hard to exceed that performance; moreover, diminishing returns kick in -- getting from 10 MB/s to 40 MB/s is a big leap (delta percent = 300%); the difference between 40 MB/s and say 50 MB/s would often be unnoticed (delta percent = 25%).
You shouldn't even need a crossover cable -- a regular cable should be fine for a direct-wire connection on (standards-compliant) gigabit NICs. Just FYI; there's nothing wrong with using a crossover cable.
The following stats illustrate that Windows client OS's can also achieve still higher performance under some conditions:
Directories processed = 1
Total data in bytes = 8,425,053,778
Elapsed time in sec. = 103.3
Files copied = 1
That's 81.6 MB/s actual file transfer performance, from Windows 2000 Pro to XP Home, from one RAID 0 array to another, using all consumer hardware, on-board NICs, on-board RAID, and a consumer switch without jumbo frames enabled.
I'm certainly not saying that I can always get this sort of speed, nor saying that RAID 0 is a good idea in general. I'm just saying that even with commonly-available consumer parts, including the OS's, such speeds are achievable for file transfers, in some cases.
For example, the next run of the same test took 110.3s, i.e. 76.4 MB/s. Bah
madwand,
when the data is cached and doesn't suffer physical disk limitations you can fill a gig pipe.
not bad.
Many times you use multiple 1000 Base-T connections to overcome this bottleneck.
when the data is cached and doesn't suffer physical disk limitations you can fill a gig pipe.
not bad.
Many times you use multiple 1000 Base-T connections to overcome this bottleneck.
I know; I've deliberately used very big files here to be fair and show actual file transfer speeds without "cheating" using the cache.
Not really, at least not for consumer/home setups. You told us this long ago, in the "Firehose" thread, explaining how 802.3ad / etc. cannot split a single TCP/IP connection across different ports. Took me a while to get it, BTW, and I've confirmed it with some teaming HW vendors.
Your point is fine for a multi-client server setups, where multiple concurrent connections can get > 1 Gb/s throughput using multiple teamed GbE NICs.
But there's no reliable teaming protocl that I know of that will send a single file any faster than a single NIC's throughput, so I try to warn consumers from getting too excited about multiple NICs and high-end switches, etc. -- they won't work as we'd like. Great for servers, not really useful for single-user scenarios.
Thanks.
You're right. Just trying to point out that in larger environments this is overcome.
Stations to station transfers cause a not seen situation in high performance computing - same L2, L3, L4 source and destination.
The protocol that can do it is MLPPP or SONET, relegated primarily to the long haul world, not LAN.
The hardware isn't built nor smart enough to perform this singular L2/L3/L4 at 1G+ speeds and isn't really a LAN application. It would have to make ASIC/bitwise decisions based on too many branches. Or just throw a 10G card onto the server.
Stations to station transfers cause a not seen situation in high performance computing - same L2, L3, L4 source and destination.
The protocol that can do it is MLPPP or SONET, relegated primarily to the long haul world, not LAN.
The hardware isn't built nor smart enough to perform this singular L2/L3/L4 at 1G+ speeds and isn't really a LAN application. It would have to make ASIC/bitwise decisions based on too many branches. Or just throw a 10G card onto the server.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter