CCNA - Bridges double the bandwidth???

Author

Bridges double the bandwidth???

Goldorak

2004-05-29, 4:04 pm

Hi,

I quote Wendell Odom - Intro - Page 240 - Paragraph 3:
"Before bridges were created, a 10BaseT network might have begun to suffer from performance problems. As described in the previous section, to improve performance, you might have added a 2-port bridge, created 2 LAN segments, doubled the bandwdth , reduced collisions, and improved performance."

- Created 2 LAN segments >> Agree
- Doubled the bandwdth >> ???
- Reduced collisions >> Agree
- Improved performance >> Agree

I have been lead to believe that bandwidth is constant ie: 10baseT is 10 mb/s.(there could be 2 or 20 devices sharing that bandwidth, the bandwidth is still 10 mb/s).
So, am I missing a crucial point or does Mr Odom mean to say that "adding a 2 port bridge to the LAN would double the throughput"?

What do you guys think?

Thank you

azimuth40

2004-05-29, 4:20 pm

A bridge can bridge independant segments. Rember everyone is listening to all traffic on any single segment

Think about using a computer as a bridge and what he is getting at makes a little more sense. Stick two 10baseT NIC's in a computer and hook nodes to each NIC. Each of the NIC's is 10Mb and as long as the traffic does not cross the bridge then you have double the bandwidth. As soon as traffic crosses back and forth across the bridge from one NIC to the other then you drop back down to 10Mb.

If most of your traffic has no need to cross the bridge then you get double the bandwidth most of the time. That was the reason that Novell originally recommended multiple NIC's in Novell servers.

Goldorak

2004-05-29, 4:38 pm

azimuth40:
Sorry didn't think it was possible to make one 10baseT NIC on a node receive data at the rate of 20mb/s? Because this is what we are talking about : one nic on each node and a 10baseT at 20mb/s thanks to the bridge down the road???
Please clarify.

Yankee

2004-05-29, 6:19 pm

I believe what the author is saying though not well phrased and makes too many assumptions, is that if you have 20 hosts on one LAN, then split them into two LANs so there are 10 hosts on each the physical BW is still 10M, but you have half as many hosts accessing it, so in a distorted sort of way you have twice the BW available. Seems a silly way to say it, but that is all I can guess that he meant.

Yankee

azimuth40

2004-05-29, 9:49 pm

Goldorak, you are correct ONE single node can not, but a bridge is bridging TWO node segments and therefore TWO NIC's. The bridge is listening to two node segments for possible work to do and therefore had better be capable of 20Mb's/sec

Again I go back to my dual NIC in a machine analogy. Just consider a simple file transfer, the system processor and hard disk are fully capable of saturating a 10Mb NIC. One system hooked to NIC A starts a file transfer and one system hooked to NIC B starts a file transfer. The maximum bandwidth of NIC A is used and the maximum bandwidth of NIC B is also used for an aggregate of 20Mb. That is because each segment is isolated from the other one.

3Com capitalizes on this with their parallel tasking II load balancing as a reason to spend more on their products. You can view their Q&A on this here.

You are able to demonstrate this anomaly even on a simple bog standard XP Pro system with two NIC's. The Networking tab in task manager will give you statistics on each of the two connections. The ability to function as a bridge is part of XP Pro. I don't know if that feature is in XP Home

Now if the system on NIC A wants to talk to a system on NIC B the maximum bandwidth is 10Mb. That is because they are no longer isolated and are bridged by the system or independent bridge. It's not difficult to visualize if you examine the transfer scenarios carefully.

By the way this same principal is used elsewhere. RAID 0 with pairs of disks on different controllers to increase bandwidth.

Goldorak

2004-05-30, 4:04 am

Hum...

Let's assume the following:
Both sketches represent a 10BASET network

Scenario Number 1
http://www.gatwick.plus.com/bridges_files/image001.gif
Above, A, B, C and D all share 10mb/s.
So each node has a theoretical throughput or 10/4 = 2.5 mb/s.

Here 2.5mb/s is not bandwidth. It is called throughput.

Scenario Number 2
http://www.gatwick.plus.com/bridges_files/image002.gif
In this scenario, we have added a bridge in the middle so A and B share 10mb/s and so do C and D.
Now each node has a throughput or 10/2 = 5 mb/s.

Again, here, 5 mb/s is not bandwidth. It is called throughput.

In scenario number one, we had 4 nodes contending with 10 mb/s.

In scenario number 2 we have put a bridge in the middle. This bridge has the effect of reducing contention by half by creating 2 broadcast domains.

Yes, scenario number 2 has double data rate but in any case, the bandwidth remains 10mb/s

The only way you can double the bandwidth is by adding a second nic on each computer not by adding a bridge in the middle.

Yankee

2004-05-30, 9:34 am

This is getting silly.... You could put a GIG NIC in the dang puter, the bandwidth would still be 10M and your throughput would still be limited to 10M.

You could put two GIG NICs and have them on the different LANs. If their destination point is the same you would still be limited to the smallest BW both paths must travel.

Yankee

azimuth40

2004-05-30, 5:00 pm

Goldorak I am about done because as Yankee states it is getting silly. You asked how Odem came up with his statement and I tried to explain the rational. I gave you a link to 3Com with a FAQ and you now change from bandwidth to throughput. Your apparent definition of throughput is not correct for several reasons.

Your original question as it relates to ODEM was only about bandwidth which is the maximum transfer possible and not the maximum transfer sustainable. Ethernet by its very nature can not sustain its bandwidth and throughput is not just simple division factor. That would be like people that believe that their hard disk really transfers 133MB/sec all the time when the trough put has to do with seek times, angular velocity, bit density per square inch, internal buffer size and many other factors.

Also while you are trying to split things in your diagram did you consider half-duplex versus full-duplex NIC's. Research that angle and I think you will find that you are still missing the point. What you have suggested at its base level is also more akin to tokenring and not Ethernet.

If all nodes are transferring at the same time Ethernet throughput will never be a simple division factor because of excessive collisions. I did not include that as part of the original discussion because I originally thought your "troughput" at the end as a typo as the question at the beginning said bandwidth. While throughput increases, it never doubles when you add two of anything because of maintenance overhead. Dual processors are not twice as fast as single processors either.

Finally the reason for splitting a network into isolated segments is to allow simultaneous transfers on each segment when bridging is not needed at the moment. You do realize that bridging occurs at the packet level by looking at the MAC address and is not a continuous connection right? That is the reason that switches replaced hubs even though until recently they had a substantial premium cost. It is also the reason that I used a computer with two NIC's performing the bridging function as an example.

I am done here, it was not my intent to start an argument. With more real world experience it will become clear to you. I suggest researching bridges more at the hardware level and understand why large networks often bridge with computers to gain add on benefits like using STP, multiple bridges per box and firewalling.

Sexy Lexy

2004-05-30, 7:17 pm

This question has been answered, thank you azimuth40 and Yankee for your input.

Goldorak, the explaination given explains the situation that you presented.

Don't take their answers the wrong way, they are trying to help.

Sponsored Links