Information Technology Reference
In-Depth Information
•
TCP connections are slow because of cells being discarded in the ATM cloud. This results in IP
packets being discarded and in a high number of retransmissions. TCP itself believes that this is
because of congestion and will try to lower its transmitting window, resulting in a very slow TCP
connection. Of course, this affects all TCP-based protocols, including Telnet and FTP.
•
Large IP packets tend to fail, but small packets cross the ATM network with no problem. This is
again because of cells being discarded.
Let's concentrate on this second symptom, to help us detect the problem. Say that for every 100 cells
transmitted back to back by the source router, the cloud will discard the last one because of policing.
This means that if a
ping
has a data part of 100 bytes, 3 ATM cells will be needed to send it (because 3
´
48 bytes will be needed to contain the ICMP echo request). In practice, this means that the 33 first
ping
s will succeed (more precisely, the first 99 cells will be seen within contract by the provider), but
the 34th one will fail because one of its cells will be discarded.
Assuming that we keep the same setup and use 1500-byte packets instead of small ICMP echos (
ping
s),
we will need 32 cells to transmit each large packet (32
48 = 1536 bytes, the smallest multiple of 48
above the packet size). If the network discards 1 cell out of 100, about one packet out of three or four
will be discarded. Raising the packet size is then a simple and efficient way to prove that you have a
policing issue.
In practice, you can generate large
ping
s from the router itself:
Medina#
ping
Protocol [ip]:
Target IP address: 10.2.1.2
Repeat count [5]: 100
Datagram size [100]: 1500
Timeout in seconds [2]: 2
Extended commands [n]:
Sweep range of sizes [n]:
Type escape sequence to abort.
Sending 100, 1500-byte ICMP Echos to 10.2.1.2, timeout is 2 seconds:
!!!.!!.!!!.!!.!!!.!!.!!!.!!.!!!.!!.!!!.!!.!!!.!!.!!!.!!.!!!.!!.!!!.!!.!!!.!!.!!!.!!.!!!.!!
.!!!.!!.!
Success rate is 72 percent (72/100)
If the real problem is related to policing, doing the same test with larger packets will generate a totally
different result:
Medina#
ping
Protocol [ip]:
Target IP address: 10.2.1.2
Repeat count [5]: 100
Datagram size [100]: 3000
Timeout in seconds [2]: 2
Extended commands [n]:
Sweep range of sizes [n]:
Type escape sequence to abort.
Sending 100, 3000-byte ICMP Echos to 10.2.1.2, timeout is 2 seconds:
!.!.!..!.!.!..!.!..!.!...!..!.!.!..!.!.!.!.!.!.!..!..!.!...!..!.!.!..!.!.!..!.!.!..!.!..!.
!.!.!..!..!
Success rate is 42 percent (42/100)
´
If, after running those tests, you conclude that you are suffering from a policing issue, contact your ATM
provider immediately and check the following points:
•
Is the provider indeed discarding cells? The provider
must
be capable of telling you this.
•
If so, for what specific reason is this happening? The answer will usually be policing, but sometimes
the network is simply congested.
If the reason is policing, then what are the traffic parameters? Do they match with the settings on the
router?
