1 Gigibit Ethernet - Slow?

[Brandon Brown]

I bought 2 Dlink DGE-530T 1000 Mbps (1 Gbps) Ethernet cards to replace my
10/100 Mbps. To test it, I installed on 2 WinXP computers and linked with a
Cat 5
crossover cable. The icons on the system tray showed "Speed: 1 Gbps" on both
computers

Then, I tried to download some files form one computer to another using both
Share drive and Webserver, the result was so disappointed... on average
it's about 3 MB/sec which is about 24 Mbps <-- 24% of 100 Mbps!!! My old
10/100 Mbps could do the same???!!!! I assume 1000 Mbps would give me at
least 200 Mbps in trasfer speed! Am I correct?

Did I do something wrong? Is there something I need to setup in Windows,
registry or some settings in driver to
get faster speed? please help.

Thanks

 

[Mike Walsh]

The bottleneck is probably one of your hard drives. Try transferring a large (e.g. 500 MB) unfragmented file. You might need faster drives, maybe a new RAID0 setup.

 

[Cuzman]

" I assume 1000 Mbps would give me at least 200 Mbps in trasfer speed! Am I
correct? "


If you put a huge exhaust on a shit car, it doesn't become a Ferrari. You
are always limited by the slowest link in the chain.

The 32-bit PCI bus bandwidth is limited to 127.2Mbps, so 200Mbps is
unrealistic. At the moment, gigabit ethernet (GbE) only tends to show real
improvements for network backbones with 64-bit PCI 2.1 bus speeds.
PCI-Express motherboards with Serial-ATA RAID-0 configurations will improve
GbE speeds for workstations over the coming year, but you still might not
reach 200Mbps with that.

 

[*Vanguard*]

Brandon Brown said in

I bought 2 Dlink DGE-530T 1000 Mbps (1 Gbps) Ethernet cards to
replace my 10/100 Mbps. To test it, I installed on 2 WinXP computers
and linked with a Cat 5
crossover cable. The icons on the system tray showed "Speed: 1 Gbps"
on both computers

Then, I tried to download some files form one computer to another
using both Share drive and Webserver, the result was so
disappointed... on average it's about 3 MB/sec which is about 24 Mbps
<-- 24% of 100 Mbps!!! My old 10/100 Mbps could do the same???!!!!
I assume 1000 Mbps would give me at least 200 Mbps in trasfer speed!
Am I correct?

Did I do something wrong? Is there something I need to setup in
Windows, registry or some settings in driver to
get faster speed? please help.

Thanks

What happens if you setup a ramdrive on each host and transfer files
between those?

 

[kony]

" I assume 1000 Mbps would give me at least 200 Mbps in trasfer speed! Am I
correct? "


If you put a huge exhaust on a shit car, it doesn't become a Ferrari. You
are always limited by the slowest link in the chain.

The 32-bit PCI bus bandwidth is limited to 127.2Mbps, so 200Mbps is
unrealistic. At the moment, gigabit ethernet (GbE) only tends to show real
improvements for network backbones with 64-bit PCI 2.1 bus speeds.
PCI-Express motherboards with Serial-ATA RAID-0 configurations will improve
GbE speeds for workstations over the coming year, but you still might not
reach 200Mbps with that.

Nope, PC bus limit is around 127 MegaBYTES = 1000 MegaBITS.
In a PC the PCI bus is one of the bottlenecks, but nowhere near signifiant
enough to prevent 200Mbps the OP expected.

 

[Bubba]

Brandon Brown's log on stardate 22 svi 2004

Then, I tried to download some files form one computer to another
using both Share drive and Webserver, the result was so
disappointed... on average it's about 3 MB/sec which is about 24 Mbps
<-- 24% of 100 Mbps!!! My old 10/100 Mbps could do the same???!!!!
I assume 1000 Mbps would give me at least 200 Mbps in trasfer speed!
Am I correct?

Cabel? Length?

 

[Brandon Brown]

Cat 5 cable, less than 10 feet.. as I said in first message, I bought the
cards just to test it to see if I want to invest in the 1 Gbps switch
(switch is expensive)

 

[Brandon Brown]

That's a VERY good idea... I should try that.

Brandon Brown said in



What happens if you setup a ramdrive on each host and transfer files
between those?


--
____________________________________________________________
*** Post replies to newsgroup. Share with others.
*** Email: domain = ".com" and append "=NEWS=" to Subject.
____________________________________________________________

 

[kony]

I bought 2 Dlink DGE-530T 1000 Mbps (1 Gbps) Ethernet cards to replace my
10/100 Mbps. To test it, I installed on 2 WinXP computers and linked with a
Cat 5
crossover cable. The icons on the system tray showed "Speed: 1 Gbps" on both
computers

Then, I tried to download some files form one computer to another using both
Share drive and Webserver, the result was so disappointed... on average
it's about 3 MB/sec which is about 24 Mbps <-- 24% of 100 Mbps!!! My old
10/100 Mbps could do the same???!!!! I assume 1000 Mbps would give me at
least 200 Mbps in trasfer speed! Am I correct?

Did I do something wrong? Is there something I need to setup in Windows,
registry or some settings in driver to
get faster speed? please help.

Thanks

Is your crossover cable fully wired for Gb, or only Mb?

Gb uses all 4 pairs of wires,
http://logout.sh/computers/net/gigabit/

I don't know if such a cable problem would result or change the "1 Gbps"
icon you see or not.. in a perfect world it would, but in this one???

 

[DaveW]

GB ethernet will run over Cat. 5 cable no faster than 100 Mbps. GB ethernet
requires Cat. 6 cable.

 

[kony]

GB ethernet will run over Cat. 5 cable no faster than 100 Mbps. GB ethernet
requires Cat. 6 cable.

Although CAT5 only requires 2 pair of wires, "almost" every CAT5, 5e cable
does have 4 pair, certainly 5e is acceptable for use with Gb unless cable
runs are quite long. CAT5 "can" work too, shouldn't be restricting speed
to 100Mb level. However, in this particular instance it could be
significant that the cable is a crossover, if only two pair are crossed.

 

[Trent©]

That's a VERY good idea... I should try that.

You better have a fast stop watch...unless they've radically changed
the limitations of ram drives!


Have a nice week...

Trent©

Follow Joan Rivers' example --- get pre-embalmed!

 

[Trent©]

I bought 2 Dlink DGE-530T 1000 Mbps (1 Gbps) Ethernet cards to replace my
10/100 Mbps. To test it, I installed on 2 WinXP computers and linked with a
Cat 5
crossover cable. The icons on the system tray showed "Speed: 1 Gbps" on both
computers

Then, I tried to download some files form one computer to another using both
Share drive and Webserver, the result was so disappointed... on average
it's about 3 MB/sec which is about 24 Mbps <-- 24% of 100 Mbps!!! My old
10/100 Mbps could do the same???!!!! I assume 1000 Mbps would give me at
least 200 Mbps in trasfer speed! Am I correct?

Did I do something wrong? Is there something I need to setup in Windows,
registry or some settings in driver to
get faster speed? please help.

Purchase a comparable router to go with the setup.


Have a nice week...

Trent©

Follow Joan Rivers' example --- get pre-embalmed!

 

[*Vanguard*]

Trent© said in news:[email protected]:

You better have a fast stop watch...unless they've radically changed
the limitations of ram drives!

So transfer the same file a hundred or a thousand times for a repeated
overwrite.

 

[kony]

Trent© said in news:[email protected]:

So transfer the same file a hundred or a thousand times for a repeated
overwrite.

Huh? What's wrong with just transferring a 300MB single file? I suppose
he'd have to hunt down a 3rd party ramdrive driver though.

 

[*Vanguard*]

kony said in news:[email protected]:

Huh? What's wrong with just transferring a 300MB single file? I
suppose he'd have to hunt down a 3rd party ramdrive driver though.

I assumed Trent was thinking the 64MB file on a 64MB ramdrive. Isn't
the max size for, say, the Microsoft ramdrive and many other freebie
ramdrive utilities about 64MB (I don't use a ramdrive)? A 64MB transfer
might finish too quickly that you wouldn't get enough time to make an
accurate measurement to compute the transfer rate. So I figure you
could transfer the file as many times as it takes to get about 10, or
more, seconds worth of traffic. Anything under a second would have too
much variance to make a good measurement; a difference of half a second
for a 1 second max measure is a lot more variance than half a second for
a 10 second measure, and the longer your total time the more accurate
the measurement. The OP said he was getting around 3MBps so, yeah, it
would take 20 seconds to transfer a 64MB file but that's under whatever
is currently throttling his throughput and will probably not exist when
using a ramdrive. If the OP got just half of the 127MBps from the PCI
bus when using a ramdrive, his transfer of a 64MB file would be under a
second which is way too short for an accurate measurement.

The OP wouldn't need 1000 copy commands in a .bat file. He could just
run "FOR /L %variable IN (start,step,end) DO command
[command-parameters]" from a command prompt. The idea is to generate
enough traffic over a long enough time period to provide a reasonable
measurement. I'd just keep upping the end value until the transfer of
all copies took around 10 seconds. However, it would be difficult to
keep track of how many times the FOR look had looped: couting hundreds
or thousands of lines of output would be too arduous, and not possible
unless the DOS window were configured to enlarge its buffer to thousands
of lines. Run the following:

for /l %i in (1,1,N) do (
copy <src> <dest> > nul
echo Copy %1 completed.
)

You can put this in a .bat file. You can also run this from the command
prompt: you get a "More?" prompt until you close the compound statement
(i.e., enter the closing parenthesis). I would start out with N = 10
and move up from there, like to 100, multiples of 100, 1000, multiples
of 1000, until I got 10 seconds worth of traffic. You know the size of
the file, how many times it got transferred, and the time it took.

 

[Geir Klemetsen]

Trent© said in news:[email protected]:

So transfer the same file a hundred or a thousand times for a repeated
overwrite.

Set up a ramdrive on both machines.
Write down the folder path for the folder/file you want to copy from/to and
write it down, and sent as reply to the newsgroup, and I can make you a
*.vbs -file that copy the file(s)
and measures the time it takes. It's a very simple task, takes me about 5
min to write.
That will make it possible to get accurate measurment even if the time it
takes is less than 1/100 second.
And, of course I'm courious what the result will be.

 

[Noozer]

I bought 2 Dlink DGE-530T 1000 Mbps (1 Gbps) Ethernet cards to replace my
10/100 Mbps. To test it, I installed on 2 WinXP computers and linked with a
Cat 5
crossover cable. The icons on the system tray showed "Speed: 1 Gbps" on both
computers

Then, I tried to download some files form one computer to another using both
Share drive and Webserver, the result was so disappointed... on average
it's about 3 MB/sec which is about 24 Mbps <-- 24% of 100 Mbps!!! My old
10/100 Mbps could do the same???!!!! I assume 1000 Mbps would give me at
least 200 Mbps in trasfer speed! Am I correct?

I get peaks of up to 25%, but that's with very large files. The bottleneck
at this point is the speed of the IDE drive or PCI bus.

 

[Noozer]

That's a VERY good idea... I should try that.

Huh? What's wrong with just transferring a 300MB single file? I suppose
he'd have to hunt down a 3rd party ramdrive driver though.

Just remember guys... transferring a bunch of little files won't give you
the same throughput as one single large file - even from a RAMDISK. To
really see the different make 1000 1K files (or a BAT to copy the same one
1000 times) and then make a single 1000K file. Copy each across the LAN
(even at 100mb) and watch the Network tab of the Taskmanager and see what
percentage of network utilization you get.

Best bet, get some benchmark software and it will generate the data as fast
as the CPU can go...

RAMDISK would work if you could greate something as big as a gigabyte or
two... Theoretically, a one gigabyte file would only take 8 seconds to copy
(8 bits to a byte) over a gigabit network (it'll never happen that fast
though).

....it's all moot anyhow as in real life the files will only go as fast as
the hard drive can move them.

 

[CBFalconer]

.... snip ...

Just remember guys... transferring a bunch of little files won't
give you the same throughput as one single large file - even from
a RAMDISK. To really see the different make 1000 1K files (or a
BAT to copy the same one 1000 times) and then make a single 1000K
file. Copy each across the LAN (even at 100mb) and watch the
Network tab of the Taskmanager and see what percentage of network
utilization you get.

Better yet, use the virtual memory abilities to generate a really
big RAMDISK, with the VM on a networked drive. Then run your
tests. <g, d, & r>