Transfer rate on LAN?

I have a LAN in my home and when I transfer files from one machine to
another I get transfer rates of between 900K and 1M per second. This seems
to be the case no matter what combination of ethernet cards and cables that
I use. Is that normal?

My setup: I have a linux box with 2 ethernet cards serving as DHCP and
caching DNS server. Eth0 is a 10 mb card connected to my DSL modem. The
other card is 100 mb card that is connected to a hub. Will linux talk
different speeds to the different cards? Maybe the whole network is slow
because of that one card? I figured that the DSL modem is way slower than
the NIC anyway so it wouldn't matter.

On 23 Jul 2005 18:42:10 CST, John Heim <(E-Mail Removed)> wrote:
>
> My setup: I have a linux box with 2 ethernet cards serving as DHCP and
> caching DNS server. Eth0 is a 10 mb card connected to my DSL modem. The
> other card is 100 mb card that is connected to a hub.
Sounds okay to me.
> Will linux talk
> different speeds to the different cards?
Yes it will, unless you're benchmarking via the modem

> Maybe the whole network is slow
> because of that one card?
No, but what sort of transfer test are you using?

# mount /home/public/
# time dd if=/dev/zero of=/home/public/zeroes bs=1M count=100
100+0 records in
100+0 records out

real 0m11.953s
user 0m0.001s
sys 0m0.329s
#
That's over 8MB/s from a AMD K7 2600+ --> pII/400 box with quality
NICs (onboard via-rhine to Intel pro/100 - e100) via NFS transport.

Following sequence shows 2 Linux boxen talking to/from winxp with SMB

root@sempro:~# mount /mnt/linux/
root@sempro:~# time dd if=/dev/zero of=/mnt/linux/zeroe3 bs=1M count=100
100+0 records in
100+0 records out

real 0m15.401s <<== 6.5MB/S
user 0m0.000s
sys 0m0.538s
root@sempro:~# time dd of=/dev/null if=/mnt/linux/zeroe2 bs=1M count=100
100+0 records in
100+0 records out

real 0m15.433s
user 0m0.001s
sys 0m0.534s

peetoo:~$ mount /mnt/linux/
peetoo:~$ time dd if=/dev/zero of=/mnt/linux/zeroe2 bs=1M count=100
100+0 records in
100+0 records out

real 0m18.163s <<== 5.5MB/s
user 0m0.004s
sys 0m2.585s
peetoo:~$ time dd of=/dev/null if=/mnt/linux/zeroe3 bs=1M count=100
100+0 records in
100+0 records out

real 0m18.018s
user 0m0.005s
sys 0m1.152s
peetoo:~$

> I figured that the DSL modem is way slower than
> the NIC anyway so it wouldn't matter.

You are right, it does not matter. Check NICs you are using, even a
p233/mmx firewall box from/to peetoo box does 3.5MB/s / 3.9MB/s over
NFS with pro/100 <-> pro/100, like yours, the firewall has an old
10Mbps NIC going to the modem.

Grant.

G_r_a_n_t_@dodo.com.au wrote in
>
>> Maybe the whole network is
>> slow
>> because of that one card?
> No, but what sort of transfer test are you using?

Primarily scp. When you scp a file, it gives you the transfer rate. But
also HTTP and FTP. If I transfer a file from my server to my Windows
machine, I get similar speeds, about 1M/sec. I commonly wget a file to
the linux server and then copy it to other machines on my LAN.

I have 2 Windows machines, 2 linux machines, and an Apple Macintosh
connected through the hub and they all get the same rate (approximately)
when getting files from the server.

The machine I mainly run tests on is about 6 feet from the server.
There's a 3 foot cable running from the linux server to the hub and then
an 8 foot cable running to the other machine. But there is another
machine on my network that is on another floor and there is a cable
running through the walls and ceilings over to it. Is the speed of the
LAN dependent upon the speed to the slowest connection? I mean, will the
server talk different speeds to different machines on eth1?

To make sure all of your machines connected to the hub all have 100Mbit
card and the hub supports 100Mbit correct?

If so, the average that you should usually see for server to computer
transfer is aproximatly 5-6 MiB a second, However remember that hub is
an dumb device, in that it broadcast the information to all interface
and it increases the network traffic so it probably would affect your
speed, if you get an switch it should improve.

-paul

John Heim wrote:
> G_r_a_n_t_@dodo.com.au wrote in
>
>>> Maybe the whole network is
>>> slow
>>>because of that one card?
>>
>>No, but what sort of transfer test are you using?
>
>
> Primarily scp. When you scp a file, it gives you the transfer rate. But
> also HTTP and FTP. If I transfer a file from my server to my Windows
> machine, I get similar speeds, about 1M/sec. I commonly wget a file to
> the linux server and then copy it to other machines on my LAN.
>
> I have 2 Windows machines, 2 linux machines, and an Apple Macintosh
> connected through the hub and they all get the same rate (approximately)
> when getting files from the server.
>
> The machine I mainly run tests on is about 6 feet from the server.
> There's a 3 foot cable running from the linux server to the hub and then
> an 8 foot cable running to the other machine. But there is another
> machine on my network that is on another floor and there is a cable
> running through the walls and ceilings over to it. Is the speed of the
> LAN dependent upon the speed to the slowest connection? I mean, will the
> server talk different speeds to different machines on eth1?
>
>
>

On 23 Jul 2005 21:09:21 CST, John Heim <(E-Mail Removed)> wrote:
> > No, but what sort of transfer test are you using?
>
> Primarily scp. When you scp a file, it gives you the transfer rate. But
Not good, scp is transferring encrypted files via ssh, so you
are measuring CPU processing overhead )

> also HTTP and FTP. If I transfer a file from my server to my Windows
> machine, I get similar speeds, about 1M/sec. I commonly wget a file to
> the linux server and then copy it to other machines on my LAN.

Compare NFS to scp file transfer:

grant@peetoo:~$ time cp /home/share/linux-2.6/linux-2.6.12.tar.bz2 .

real 0m7.195s
user 0m0.005s
sys 0m0.789s
grant@peetoo:~$ time scp deltree:/home/share/linux-2.6/linux-2.6.12.tar.bz2 .
....
linux-2.6.12.tar.bz2 100% 36MB 1.3MB/s 00:28

real 0m34.015s
user 0m6.242s
sys 0m1.729s

So NFS ~4 times faster (subtract user time, scp needed authentication)
than scp for same file transfer ) But I'm down towards your ~1MB/s
rate with scp.

Grant.

John Heim <(E-Mail Removed)> wrote in
news:Xns969CBE6BF4591johnheimjohnheimcom@216.170.1 53.136:

> I have a LAN in my home and when I transfer files from one machine to
> another I get transfer rates of between 900K and 1M per second. This
> seems to be the case no matter what combination of ethernet cards and
> cables that I use. Is that normal?

Is that bits or bytes per second? If million bytes per second then that
would be about right for 10 m bit per second ethernet. See below.

> My setup: I have a linux box with 2 ethernet cards serving as DHCP and
> caching DNS server. Eth0 is a 10 mb card connected to my DSL modem.
> The other card is 100 mb card that is connected to a hub.

What speed is the hub capable of? If it is only 10mb then your 100 mb
cards will negotiate down if they are set to auto negotiation. You would
be better off with an ethernet switch than a hub anyway.

Klazmon.

Just what I was thinking. With a hub, how is it not going to be limited
by the speed of its slowest connection. If you plug a 10mb connection
into the HUB, all connections will be effectively 10mb. Plus hubs have
a lot of collisions and are 1/2 duplex. It will still physically link
at 100Mb, but throughput has to be limited by the slowest computer. And
that limit is probably going to come in the form of collisions.

switch is way better. I dont think you can even find hubs anymore...

CL


Paul wrote:
> To make sure all of your machines connected to the hub all have 100Mbit
> card and the hub supports 100Mbit correct?
>
> If so, the average that you should usually see for server to computer
> transfer is aproximatly 5-6 MiB a second, However remember that hub is
> an dumb device, in that it broadcast the information to all interface
> and it increases the network traffic so it probably would affect your
> speed, if you get an switch it should improve.
>
> -paul
>
> John Heim wrote:
>
>>G_r_a_n_t_@dodo.com.au wrote in
>>
>>
>>>> Maybe the whole network is
>>>> slow
>>>>because of that one card?
>>>
>>>No, but what sort of transfer test are you using?
>>
>>
>>Primarily scp. When you scp a file, it gives you the transfer rate. But
>>also HTTP and FTP. If I transfer a file from my server to my Windows
>>machine, I get similar speeds, about 1M/sec. I commonly wget a file to
>>the linux server and then copy it to other machines on my LAN.
>>
>>I have 2 Windows machines, 2 linux machines, and an Apple Macintosh
>>connected through the hub and they all get the same rate (approximately)
>>when getting files from the server.
>>
>>The machine I mainly run tests on is about 6 feet from the server.
>>There's a 3 foot cable running from the linux server to the hub and then
>>an 8 foot cable running to the other machine. But there is another
>>machine on my network that is on another floor and there is a cable
>>running through the walls and ceilings over to it. Is the speed of the
>>LAN dependent upon the speed to the slowest connection? I mean, will the
>>server talk different speeds to different machines on eth1?
>>
>>
>>


--
Respectfully,


CL Gilbert

Llanzlan Klazmon wrote:
> John Heim <(E-Mail Removed)> wrote in
> news:Xns969CBE6BF4591johnheimjohnheimcom@216.170.1 53.136:
>
>
>>I have a LAN in my home and when I transfer files from one machine to
>>another I get transfer rates of between 900K and 1M per second. This
>>seems to be the case no matter what combination of ethernet cards and
>>cables that I use. Is that normal?
>
>
> Is that bits or bytes per second? If million bytes per second then that
> would be about right for 10 m bit per second ethernet. See below.
>
>
>>My setup: I have a linux box with 2 ethernet cards serving as DHCP and
>>caching DNS server. Eth0 is a 10 mb card connected to my DSL modem.
>>The other card is 100 mb card that is connected to a hub.
>
>
> What speed is the hub capable of? If it is only 10mb then your 100 mb
> cards will negotiate down if they are set to auto negotiation. You would
> be better off with an ethernet switch than a hub anyway.
>
> Klazmon.

I dont think they will negotiate down. Physical link will remain at
100mb. But the throughput will be limited.

--
Respectfully,


CL Gilbert

CL (dnoyeB) Gilbert wrote:

> With a hub, how is it not going to be limited
> by the speed of its slowest connection. If you plug a 10mb connection
> into the HUB, all connections will be effectively 10mb.

There are (were?) some switched hubs available, which consisted of two hubs,
one at 10 Mb, the other at 100 Mb, connected via a bridge or switch.

James Knott wrote:
> CL (dnoyeB) Gilbert wrote:
>
>
>>With a hub, how is it not going to be limited
>>by the speed of its slowest connection. If you plug a 10mb connection
>>into the HUB, all connections will be effectively 10mb.
>
>
> There are (were?) some switched hubs available, which consisted of two hubs,
> one at 10 Mb, the other at 100 Mb, connected via a bridge or switch.
>

Even worse, some hubs consisted of a separate 10Mb and 100Mb hub with no
connection between them (I vaguely remember some 3Com model, where an
xxx1 did have this setup, and the more expensive xxx2 had a bridge
between the hubs)

J.