True Gigabit Transfer Speed

[anthonymeluso]

Heres the thing I don't understand a 100mb Base-T Ethernet will usually result in a transfer rate of 11mbs on our network. However when testing a 1000mb Base-T test network I would receive speeds up to 40mbs.

So what gives? Shouldn't 1000mb Base-T connections be close to 10x faster than 100mb Base-T connections. I mean there is an improvement but nothing too dramatic. Should I be using CAT 6 wire rather than CAT 5e?

 

[KiscoKid]

How are you verifying this? With packet generators attached end-to-end or by running download tests from some Internet site (or something similar)?

The reason I ask is that using remote servers hosted on the Internet for example introduces a lot of variables that can greatly influence the speed you see.

Such variables include: server congestion, packet loss, WAN bottlenecks, QoS, TCP windowing and more.

 

[jimbopalmer]

Are you sure your servers can generate more than 11 mbit? It would need to to be from a PCI-X or PCI-E RAID to a PCI-X or PCI-E RAID, as the PCI Bus is a minor limit to speed but IDE drives are typically not capable of 20 mbit without RAID. 'Test' rigs typically do not test the drives, just the network.

It is possible the servers are CPU bound, ('server' NICs can help) you may wish to watch CPU utilization. I would suspect that is why the test rig is limited to 40 mbit, if the production rigs have more to think about, it will slow them down.

I tried to remain child-like, all I acheived was childish.

 

[anthonymeluso]

I'm using a program called Netcps. I'm just connecting two workstations together with the same CAT 5 wiring that we used throughout the building. The workstations both have Gigabit adapters on them and both show that they are connected at 1 Gigabit as well.

I'm pushing through 3GB of data for the test and I am noticing speeds around 40mbs. What speeds should I be noticing? AKA if you guys run the test on your network what speeds do you see?

Thanks.

 

[MaCinTof]

Theoreticly you should have something more like 800Mo/s for the gigabit Ethernet.
May be the limitation come from your hard drive or your workstations.
Have you checked that ?

 

[anthonymeluso]

These workstations are brand new Dells with Intel Core 2 Duos in them so I don't think its the computers. I'm going to try tomorrow with a freshly crimped Cat 5e cable and see what I get.

Thanks again.

 

[pansophic]

Have you tried this with a Null Ethernet cable? I couldn't find any specs on the Netcps in terms of its ability to handle anything remotely approaching gigabit speeds.

Note that winsock seems to take quite some CPU resources on high-speed connections. I got ~4 MB/s max on a Intel Pentium 133, and ~7 MB/s on an AMD K6 200, using a Intel Intel Pro 200 server with NT 4.0 Server as the other host. I got a report that two Pentium 200 servers running NT 4.0 server had 10 MB/s on a direct connection (server to server, no hub or switch).

From all appearances, this performance was on a 100 Mbps interface.

I suspect that you are reaching the capabilities of the WSA interface that the author used to implement the socket communications. And to defend the author, I have to say that he claims that he wrote the application very quickly as a test for his own purposes.

According to the website and from looking at the source code, there isn't any disk access, so the application should not be disk I/O bound. It should be either CPU bound, NIC/driver bound or network bound. And it doesn't seem to network bound, so I would be looking at either the CPU or driver.

And you should be able to look at interface statistics for your switch if it is managed to see if you are receiving errored packets. If not, then it is probably not your cable.


pansophic

 

[jimbopalmer]

I am skeptical you will get gig speeds over home built cables.

I tried to remain child-like, all I acheived was childish.

 

[anthonymeluso]

What other type of cable am I suppose to use? This is 3m Cat 5e cable.

Can anyone recommend any programs to test network speed?

Thanks!

 

[tomvear]

what about the hdd attached to these dells??? im assuming theyre SATA right??? 40m/bit isnt bad for a gigabit link between two desktops

 

[anthonymeluso]

So these numbers I got aren't that bad then. I just don't understand that if everyone is crazy about these speeds when they don't come half way yet. I mean 100mb-T really does transfer at 10mb/s.

I'm starting to think this is just some touting of new tech that isn't fully developed yet. I mean yea 100mb/s is possible but in a special configured environment with fast proccesors, hdd, and expensive nics.

 

[jimbopalmer]

I just don't understand that if everyone is crazy about these speeds when they don't come half way yet. I mean 100mb-T really does transfer at 10mb/s." And gig really does gig. and 10 gig really does 10 gig, IF you have a data source and data receiver capable of generating that much data. A single PC with a single hard drive is not that source/receiver.



I tried to remain child-like, all I acheived was childish.