I am experiencing significant collisions on a 100BaseT network when just one Wintel computer on the lan is recieving packets from a NT Server (4.0). As more computers are switched on more collisions occur. The nic on the server is set to auto the same as the workststion. I have connected the server to the 100baseT hub, workstation on the same hub, when only these two machine are on and communicating - still the same problem. I have set the server nic to 100 half duplex likewise the workstation every other device on the network turned of (bar the active equipment) - still the same problem. The server card is not chattering nor is the card on the workstation. I have swapped hubs (twice), still no resolution. Any help would be appreciate
Tek-Tips is the largest IT community on the Internet today!
Members share and learn making Tek-Tips Forums the best source of peer-reviewed technical information on the Internet!
-
Congratulations SkipVought on being selected by the Tek-Tips community for having the most helpful posts in the forums last week. Way to Go!
Network Collisions 3
- Thread starter rewas
- Start date
cshortreed
Programmer
The only time I have seen this problem is with a bad cable. Although I did go through a whole stint of collisions when I set my HP Netserver on the switch and the switch had to be specially configured to the HPTX adapter in the server (apparently it doesn't auto-negotiate very well).
kest
Technical User
in may own experience one will always get collision on a 10baseT network (ethernet solution). In an ethernet environement one will alway get collosion that is the nature of ethernet. but collision shoulf not be so grate that it bring down your network ( that traffic cannot get througt)
-
2
- #4
Having collisions on a switched LAN is bad. I would put the NICs into a manual locked speed and run some checks again. If you have it locked at 100 full and still have collisions ( high rate) then take it 10 half and see what happens. If it drops, then either you have bad cabling ( likely) or bad NICs. ( My 3coms hates to auto) The cabling can be bad but until you start pushing it hard, it may not show up unless its VERY bad.
-
1
- #5
The way i see it.....all depends on yer definition of "experiencing significant collisions ".
One of the benifits of using some swithes is the ability to segment the network into different 'collision domains'. In theory, with a Layer-3 switch config.ed properly...each port is designated as a it's own collision domain. A [basic] hub on the other hand (not intending to insult yer intelligence what-so-ever...just trying to give a cleaar *.tiff) operates at Layer-2. Unlike a switch...that 'routes' frames out SPECIFIC ports (hence: seprate collision domains) a hub takes ALL data at it's input...and sends it out ALL the outputs. What this means is that every machine tied to that hub is in the same collision domain. Every packet generated on that hub...or
'sent out' that hub...goes to EVERY machine attached.
[half-duplex] Ethernet was engineered in such a way that collisions are normal, and EXPECTED to be there. If there was an Ether net...w/ <50 users on a segment...and there were no collisions.....arguably...something is VERY wrong. One thing you should understand about 1/2 Dup Ether...is that it operates most efficently (counting in the expected collisions) at = or >60% of available bandwidth utilization. My understanding is that any bandwidth utilization <60%results in oversubscription...then some severe media-access realated issues can take place.
.5 Ether is based on CSMA/CD Carrier Sense Mutipul Access with Collision Detection. Under normal circumstances, at some point, 2, lets say, workstations will attempt to access the physical medium upon which the data transfers (layer 1, or, typically, Cat 5 cable w/RJ 45 terminations). When there is a collision, both stations stop transmiting and wait a random amount of time, then retransmit. If it is just that one wintel...and the server,,,,
tell U what--
please explain to me how you know there are too many collisions...and i will elaborate frum there.
peace
[sig][/sig]
One of the benifits of using some swithes is the ability to segment the network into different 'collision domains'. In theory, with a Layer-3 switch config.ed properly...each port is designated as a it's own collision domain. A [basic] hub on the other hand (not intending to insult yer intelligence what-so-ever...just trying to give a cleaar *.tiff) operates at Layer-2. Unlike a switch...that 'routes' frames out SPECIFIC ports (hence: seprate collision domains) a hub takes ALL data at it's input...and sends it out ALL the outputs. What this means is that every machine tied to that hub is in the same collision domain. Every packet generated on that hub...or
'sent out' that hub...goes to EVERY machine attached.
[half-duplex] Ethernet was engineered in such a way that collisions are normal, and EXPECTED to be there. If there was an Ether net...w/ <50 users on a segment...and there were no collisions.....arguably...something is VERY wrong. One thing you should understand about 1/2 Dup Ether...is that it operates most efficently (counting in the expected collisions) at = or >60% of available bandwidth utilization. My understanding is that any bandwidth utilization <60%results in oversubscription...then some severe media-access realated issues can take place.
.5 Ether is based on CSMA/CD Carrier Sense Mutipul Access with Collision Detection. Under normal circumstances, at some point, 2, lets say, workstations will attempt to access the physical medium upon which the data transfers (layer 1, or, typically, Cat 5 cable w/RJ 45 terminations). When there is a collision, both stations stop transmiting and wait a random amount of time, then retransmit. If it is just that one wintel...and the server,,,,
tell U what--
please explain to me how you know there are too many collisions...and i will elaborate frum there.
peace
[sig][/sig]
Skoobe,
Wow, I don't think I have ever heard this explained any better. If you would just finish, I think I could cut and past this into a document that I can send to all the people that ask this same question. Please, it looks like you know exactly what you are talking about, finish!
Hats off to you!
[sig][/sig]
Wow, I don't think I have ever heard this explained any better. If you would just finish, I think I could cut and past this into a document that I can send to all the people that ask this same question. Please, it looks like you know exactly what you are talking about, finish!
Hats off to you!
[sig][/sig]
#%!@ RTusk.....you just made my day!!!
whell...ohakey
...both stations stop transmiting and wait a random amount of time, then retransmit. Chances are...the next time they try to transmit, the 'line' will be free from traffic, and their data will flow. If, like I mentioned before, another station trys to transmit simultainously, it will be the same previously mentioned process. Both stations will stop transmiting, and wait a random amount of time before trying again.
On your network....
If there is a few clients and the one server...and there are media access problem, you have some issues. Based on the EXACT statement I just made...my next step would be to put the server in it's own collision domain (definitly NOT on a hub w/workstations), possiblly along the lines of it's own switched port. If that didn't solve the problem I would get put a protocol anylizer on da line and check yer frames. The anylizer, especially if you use a good one...can tell you percisly how much bandwidth is being utilized....and the # of collisions...and various udder cool stuff that most people in the world dont care about.
By the wway...i give you props, rewas, as a programmer fer checkin all the .5 vs full duples issues and so fouth.
peaz
[sig][/sig]
whell...ohakey
...both stations stop transmiting and wait a random amount of time, then retransmit. Chances are...the next time they try to transmit, the 'line' will be free from traffic, and their data will flow. If, like I mentioned before, another station trys to transmit simultainously, it will be the same previously mentioned process. Both stations will stop transmiting, and wait a random amount of time before trying again.
On your network....
If there is a few clients and the one server...and there are media access problem, you have some issues. Based on the EXACT statement I just made...my next step would be to put the server in it's own collision domain (definitly NOT on a hub w/workstations), possiblly along the lines of it's own switched port. If that didn't solve the problem I would get put a protocol anylizer on da line and check yer frames. The anylizer, especially if you use a good one...can tell you percisly how much bandwidth is being utilized....and the # of collisions...and various udder cool stuff that most people in the world dont care about.
By the wway...i give you props, rewas, as a programmer fer checkin all the .5 vs full duples issues and so fouth.
peaz
[sig][/sig]
- Thread starter
- #8
Thanks for all the assistance and ideas, the problem is solved. It turned out to be the media converters which by default (for reasons only known to the manufacturer) were set to half duplex. I pulled them apart switched the jumpers and hey presto 1 100BaseT LAN working like a dream.
Regards
Rewas [sig][/sig]
Regards
Rewas [sig][/sig]
Does a hub ever pay attention to the format of the bits on the wire, or does it just repeat electrical signals. I understand how you may have read your last post from an osi model poster or something, but just think about it. Not trying to get nasty or anything.
Guest_imported
New member
- Jan 1, 1970
- 0
- 0
- 0
this site can help a lot with the basics of cabling and networking. Hope it helps you too!
whoop-z-E--->do i feel like a JA>.....
a hub is a Layer 1 device....and no, it does not 'pay attention' to the bit format, what-so-ever. All it does is repeat [to all it's outputs] what ever signal it recieves at it's input. A hub is really nothing more than a multi-port repeater, an 'unintelligent' device.
bridges and L-2 switches on the udder hand...actually read into the frames they recieve. What they at this point...is look at the MAC [aka hardware] destination address. These said devices keep a MAC address table that records the MAC address of every device connected to it. If that destination hardware address is attached to any of the switches respective ports...then the packet is forward out that port only. Hence: seperate collision domains (in an ethernet network, anyways). If the switch does not recgonize the MAC address...it broadcasts out every attached port in search if the intended reciepient. HENCE: a single broadcast domain. If the UNK host responds to the broadcast...the said device puts the MAC address into it's table of connected devices...'rembering' what port the reciepient was on..so the next time it gets a packet for said destination....it knows where to send it in such a way that it does not generate unnecessary traffic.
The thing to remeber about these L-2 devices is that they keep a MAC table so they know 'who' is connected to their ports.......similarly...a router keeps a routing table that keeps track of all attached networks. The distinction between the two is that L-2 devices 'work' on the MAC or hardware address and L-3 devices are concerned w/ the software address or IP address.
hope this clears that up for U!
peaz
sk0obz
a hub is a Layer 1 device....and no, it does not 'pay attention' to the bit format, what-so-ever. All it does is repeat [to all it's outputs] what ever signal it recieves at it's input. A hub is really nothing more than a multi-port repeater, an 'unintelligent' device.
bridges and L-2 switches on the udder hand...actually read into the frames they recieve. What they at this point...is look at the MAC [aka hardware] destination address. These said devices keep a MAC address table that records the MAC address of every device connected to it. If that destination hardware address is attached to any of the switches respective ports...then the packet is forward out that port only. Hence: seperate collision domains (in an ethernet network, anyways). If the switch does not recgonize the MAC address...it broadcasts out every attached port in search if the intended reciepient. HENCE: a single broadcast domain. If the UNK host responds to the broadcast...the said device puts the MAC address into it's table of connected devices...'rembering' what port the reciepient was on..so the next time it gets a packet for said destination....it knows where to send it in such a way that it does not generate unnecessary traffic.
The thing to remeber about these L-2 devices is that they keep a MAC table so they know 'who' is connected to their ports.......similarly...a router keeps a routing table that keeps track of all attached networks. The distinction between the two is that L-2 devices 'work' on the MAC or hardware address and L-3 devices are concerned w/ the software address or IP address.
hope this clears that up for U!
peaz
sk0obz
Guest_imported
New member
- Jan 1, 1970
- 0
- 0
- 0
I'm not a computer genius, so don't use technical jargon when answering this please...
my 10/100 hub is constantly getting collisions and i don;t know how to fix it. Do i have to have cross-over cables if i'm going through a hub?? how do i connect two computers without a hub
my 10/100 hub is constantly getting collisions and i don;t know how to fix it. Do i have to have cross-over cables if i'm going through a hub?? how do i connect two computers without a hub
A hub always suffers from collisions. What matters is how many computers, and how much traffic they pass over it. If you are connecting a few computers, they are OK. If you are connecting a lot of computers, go with layer two switching, and you will get rid of the collision problem. Most hubs have a collision light. If it is solid, you probably have a good indication that you have a problem with too many collisions. If you can see it flickering at a slow rate, you probably dont have to worry. (this without resorting to test equipment)
A good reason for going with layer two switching from the get go, is that if you get a chatery NIC (a defective nic card that sends out junk packets) it wont kill your network with collisions.
Now, the only pins that matter on a standard CAT5 cable are pins 1,2,3,6. Pins 1 and 2 being transmit + and - Pins 3, and 6 being recieve + and -. All you have to do to make a crosover cable is cross these wires over.
Normally, if you connect two hubs or switches together, you need a crosover cable. You dont if it has a port labeled MDIX (that stands for crosover). Some hubs and switches have a switch to cross the port over.
A quick and dirty way to test your crosover cables is to plug both ends into the same hub or switch. If both ports light up, then it is most likely a good cable.
From a hub or switch port to a computer or printer, use a straight through cable.
If you want to just connect two computers together, you can get rid of your hub IF you use a crosover cable.
Sorry for being so long, couldn't help it....
8^)
Phill
A good reason for going with layer two switching from the get go, is that if you get a chatery NIC (a defective nic card that sends out junk packets) it wont kill your network with collisions.
Now, the only pins that matter on a standard CAT5 cable are pins 1,2,3,6. Pins 1 and 2 being transmit + and - Pins 3, and 6 being recieve + and -. All you have to do to make a crosover cable is cross these wires over.
Normally, if you connect two hubs or switches together, you need a crosover cable. You dont if it has a port labeled MDIX (that stands for crosover). Some hubs and switches have a switch to cross the port over.
A quick and dirty way to test your crosover cables is to plug both ends into the same hub or switch. If both ports light up, then it is most likely a good cable.
From a hub or switch port to a computer or printer, use a straight through cable.
If you want to just connect two computers together, you can get rid of your hub IF you use a crosover cable.
Sorry for being so long, couldn't help it....
8^)
Phill
Sorry Phil but in CAT5, ALL pins matter from a signaling standpoint. That is why if you build a cable with only the two pair, it fails at CAT5 specs. Each twisted pair is designed to radiate a specific amount of energy and the "extra" pairs help keep that amount in spec. This is why the author of the CAT 3,4,5 spec does not recommend Shielded TWP since the shield keeps the emitted RF inside the cable which defeats the point of the twists.
If you want to read what Paul Kish (author of CAT5) has to say in all it's gory detail, try this link.
It was a good to a few months ago.
A second paper on cable EMF and such is from Anixter
I think I have these typed right
Mike S
If you want to read what Paul Kish (author of CAT5) has to say in all it's gory detail, try this link.
It was a good to a few months ago.
A second paper on cable EMF and such is from Anixter
I think I have these typed right
Mike S
Guest_imported
New member
- Jan 1, 1970
- 0
- 0
- 0
I'm having a similar problem.
Here's the system details
Linux Server:
Running Redhat 6.1 and samba with a 3com 10/100
NT Box
Windows NT 4.0 generic 10/100 network
When I try to capture video on the NT box and save it to the Linux server the bandwith is only about 2mb a second.
I checked the collision light and it's almost steadily lit.
Any ideas what to check.
Here's the system details
Linux Server:
Running Redhat 6.1 and samba with a 3com 10/100
NT Box
Windows NT 4.0 generic 10/100 network
When I try to capture video on the NT box and save it to the Linux server the bandwith is only about 2mb a second.
I checked the collision light and it's almost steadily lit.
Any ideas what to check.
- Status
Similar threads
- Locked
- Question