cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
2897
Views
8
Helpful
8
Replies

Bandwidth Issues

julian.hanzlik
Level 1
Level 1

Hi guys!

I hope this is the correct thread ..


I'm currently switching from a 100Mbit network to a 1000Mbit network. I have setted up two VLANs for testing. Wanted to see how "fast" it will be and created a 20GB file and transferred it from PC-A to PC-B. Both are in the same VLAN. The 20GB file took about 16min with a transfer rate of 25MB/s (so about 200Mbit). Then i tried it from PC-A to PC-C, because the cable between the Catalyst Switch and the SmallBusiness Switch is between 50-80metres long. There i transferred the file in 10minutes with 35MB/s (so 280MBit). 
My PCs have an Samung EVO 850 and i7 CPU. Tested the transfer with a cheap NETGEAR 5Port switch, only the unmanaged switch and the the PCs. Got there 90 MB/s and a time under 5 minutes. 
The switches say that they have an 1000 MBit/s connection and I don't know where to start debugging. 

It's not a big Deal because it three times faster than the current network but I'm still confused.

Here is the network:

1 Accepted Solution

Accepted Solutions

Hi Julian,

apologies i couldn't get on to this before,

52 people could transfer a file with 1Gbps and then they will loose bandwidth with every other user, correct?

simply one pc will not be able to utilise all your bandwidth during general file transfer in normal circumstances due to protocol restrictions and various things that i mentioned on previous reply.In other words If you were to transfer files between different hosts collectively you'd be able to utilize the 1Gbps connection at least close to 1Gbps or applications with multiple threads of transfers

Can I change something (?) that my files are transferred faster?

I wouldn't do anything.

having said that some of the  tcp parameters you should be able to adjust on windows registry or network card parameters same way that you adjusted them on iperf but this not something I would do as it may conflict with other applications.personally I never had a requirement to tweak the default settings on client pcs but i have to be clear it wouldn't be just transport layer restrictions(tcp) you are having with file transfers that'll also be the application layer restrictions which is your application protocol SMB 

when you use iperf you don't have the SMB protocol overhead but when you transfer files from PC to PC it involves SMB protocol

And why are the files slower than the test with iperf? Where is the difference?

As I mentioned above iperf doesn't have the smb protocol overhead as you would with windows file transfers and also it allows you to adjust the tcp parameters as you already noted.Windows automatically adjust these parameters for different protocols based on various network conditions that it identifies. additionally iperf works on memory so your disks are not involved(There's a parameter that you can set and test the performance along with the disks).

Is it only cause of the buffer size and window size ?

basically your performance will vary depending on your transport layer which is TCP or UPDP and then your application layer protocol which could be smb for general windows file transfers,ftp,http etc..

Cheers

Prabath

***Please rate all the useful posts***
-Prabath

View solution in original post

8 Replies 8

Hi Julian,

It'll be difficult to measure the bandwidth with a single file transfer as it depends on quite a few factors like,tcp performance,various I/O buffers,disks,cpu etc... in general 200Mbps is what i would expect with a file transfer test.  

I would use a tool like iperf to utilise the theoretical 1Gbps and simulate the available bandwidth.You'd find lots of information around about using iperf,there is also jperf which gives you a gui as well,if you are interested

You will not be able to compare the performance between sg200 and netgear unless you put PC-C and PC-A in the same vlan. intervlan routing on sg200 will impact your performance. are they on the same vlan?

Cheers,

Prabath

***Please rate all the useful posts***
-Prabath

Thanks for your reply Prabath, 

I will check iperf and test it with this tool. Thanks. But if I will never get more than those 200Mbit while transferring a file, when should I then get it? Transferring a file is in my opinion the only reason I would need this high bandwidht. The forwarding capacity is 52Gbps, this is when several people are copying big files. In my understanding, 52 people could transfer a file with 1Gbps and then they will loose bandwidth with every other user, correct?

Yes, the are on the same VLAN. All of them. 

Greets, 
Julian

EDIT: 
Tested the connection with iperf. PC-A to PC-B I got 590 Mbit/s. From PC-A to PC-C i got 625 Mbit/s. I changed the length of buffer to read or write and also the TCP window size to 512k. 
So, my only Questions are ..
- Can I change something (?) that my files are transferred faster? And why are the files slower than the test with iperf? Where is the difference? Is it only cause of the buffer size and window size ?

Hi Julian,

apologies i couldn't get on to this before,

52 people could transfer a file with 1Gbps and then they will loose bandwidth with every other user, correct?

simply one pc will not be able to utilise all your bandwidth during general file transfer in normal circumstances due to protocol restrictions and various things that i mentioned on previous reply.In other words If you were to transfer files between different hosts collectively you'd be able to utilize the 1Gbps connection at least close to 1Gbps or applications with multiple threads of transfers

Can I change something (?) that my files are transferred faster?

I wouldn't do anything.

having said that some of the  tcp parameters you should be able to adjust on windows registry or network card parameters same way that you adjusted them on iperf but this not something I would do as it may conflict with other applications.personally I never had a requirement to tweak the default settings on client pcs but i have to be clear it wouldn't be just transport layer restrictions(tcp) you are having with file transfers that'll also be the application layer restrictions which is your application protocol SMB 

when you use iperf you don't have the SMB protocol overhead but when you transfer files from PC to PC it involves SMB protocol

And why are the files slower than the test with iperf? Where is the difference?

As I mentioned above iperf doesn't have the smb protocol overhead as you would with windows file transfers and also it allows you to adjust the tcp parameters as you already noted.Windows automatically adjust these parameters for different protocols based on various network conditions that it identifies. additionally iperf works on memory so your disks are not involved(There's a parameter that you can set and test the performance along with the disks).

Is it only cause of the buffer size and window size ?

basically your performance will vary depending on your transport layer which is TCP or UPDP and then your application layer protocol which could be smb for general windows file transfers,ftp,http etc..

Cheers

Prabath

***Please rate all the useful posts***
-Prabath

-

How can I delete a comment?

Hi Prabath,

Thank you for your reply and for sharing your knowledge! It really helped me out and show me the way where I have to search for expending my knowledge. 
BUT what I really don't get is, yeah the TCP window size etc. have influence to the speed, but why isn't there a bottleneck when using the cheap NETGEAR switch ?


Greets, 
Julian

Hi Julian,

I don't have a clear answer for that but as for the switching power most of the switches now a day are rated as line rate given that chips used in them are very robust.

how did you do the test, Did you just replace the SG200 with the netgear?is there any way that you could isolate sg200 and do the tests and then place netgear in its place and do the test.

also out of the interest what model netgear as i do know that most of SG series and prosafe gigabit series use the same chips in their product lines.as an example gs200 8port and prosafe gs110/GS108 use the same broadcomchip bcm53312 what they called switch on chip,so we can compare in that point of view as well

Cheers,
Prabath

***Please rate all the useful posts***
-Prabath

I have all oportunities, because the network isn't live yet. I'll give it a complete test in an hour. I'll share the results here. 

And one hour equals one day here. 

I've tested following and I figured out, that the cable is maybe the bottleneck, because I switched to two short CAT5e cables which are completely new and got following: 

SAME VLAN → VLAN20

2960X – 2960X → 52sek 98.5MB/s
2960X – SG200-20 → 51sek 100.4 MB/s
2960X – SG200-30 → 53sek 96.6MB/s
2960X – SG200-40 → 52sek 98.5MB/s
SG200-20 – SG200-20 → 50sek 102.4 MB/s
SG200-20 – SG200-30 → 59sek 86.8 MB/s
SG200-20 – SG200-40 → 51sek 100.4 MB/s
SG200-30 – SG200-30 → 52sek 98.5MB/s
SG200-30 – SG200-40 → 57sek 89.8MB/s
SG200-40 – SG200-40 → 50sek 102.4MB/s

DIFFERENT VLAN → VLAN30 to VLAN20

SG200-20 – X → 54sek 94.8MB/s
SG200-30 – X → 53sek 96.6MB/s
SG200-40 – X → 50sek 102.4MB/s
X – X → 56sek 91.4MB/s

And as you can see, the problem is "gone". 

Thanks for posting your results and glad you've worked it out.

***Please rate all the useful posts***
-Prabath
Getting Started

Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community: