WO1997037464A1

WO1997037464A1 - Method and apparatus for modifying a delay time in a station accessing an ethernet network

Info

Publication number: WO1997037464A1
Application number: PCT/US1996/017853
Authority: WO
Inventors: Inc. Advanced Micro Devices; Mohan Kalkunte; Gopal Krishna; Jim Mangin
Original assignee: Advanced Micro Devices Inc
Priority date: 1996-03-29
Filing date: 1996-11-06
Publication date: 1997-10-09
Also published as: US5850525A

Abstract

Delay times are modified in Ethernet network devices by adding a randomized time interval generated in accordance with a propagation delay between two network stations. A server in a client-server arrangement is given priority access over clients by adding to the clients' InterPacket Gap (IPG) interval a random time delay between zero and a maximum value equal to no more than twice the cable delay between the server and the network hub. The server can access the network media after the IPG interval, whereas clients must wait the additional random time delay before accessing the media, thereby improving server throughput and overall network throughput. Collision mediation is improved by adding a randomly selected integer multiple of a propagation delay between two stations, where the integer multiplier is randomly selected from a predetermined range of integers. The randomly selected integer multiple of the propagation delay provides a second dimension of random selection to minimize subsequent collisions and minimize the occurrence of capture effects in losing stations.

Description

METHOD AND APPARATUS FOR MODIFYING A DELAY TIME IN A STATION ACCESSING AN ETHERNET NETWORK

Field of the Invention

The present invention relates to network interface devices and network access techniques on an Ethernet network. More particularly, the present invention relates to methods for attempting to access the Ethernet media.

Description of the Related Art

Local area networks use a network cable, or media, to link stations on the network. Each local area network architecture uses a media access control (MAC) enabling network interface cards at each station to share access on the media.

The Ethernet protocol ISO/IEC 8802-3 [ANSI/IEEE Std. 802.3, 1993 edition] defines a media access mechanism that permits all stations to access the network channel with equality. Each station includes an Ethernet interface card that uses carrier-sense multiple-access with collision detection (CSMA/CD) to listen for traffic on the media. Transmission by a station begins after sensing a deassertion of a receive carrier on the media, indicating no network traffic. After starting transmission, a transmitting station immediately checks to see if there has been a collision due to another station sending data at the same time. If there is a collision, both stations stop, wait a random amount of time, and retry transmission. Any station can attempt to contend for the channel by waiting for a predetermined time after the assertion of the receive carrier on the media, known as the interpacket gap

(IPG) interval. If a plurality of stations have data to send on the network, each of the stations will attempt to transmit in response to the sensed deassertion of the receive carrier on the media and after the IPG interval, resulting in a collision.

Ethernet networks mediate collisions by using a truncated binary exponential backoff algorithm, which provides a controlled pseudorandom mechanism to enforce a collision backoff interval before retransmission is attempted. According to the truncated binary exponential backoff algorithm, a station keeps track of the number of collisions (N) encountered by the station during transmission attempts. The station computes a collision backoff interval including the IPG interval and a slot time interval, and attempts retransmission after the collision backoff interval. The will attempt to transmit under the truncated binary exponential algorithm a maximum of sixteen times.

The slot time interval is calculated by selecting a random number of slot times from the range of zero to 2^N-1. For example, if the number of attempts N = 3, then the range of randomly selected time slots is (0,7); if the randomly-selected number of time slots is four, then the collision backoff interval will be equal to the IPG interval plus four slot time intervals. According to the above- identified Ethernet protocol, the maximum range of randomly selected time slots is 2¹⁰-1. The truncated binary exponential algorithm has the disadvantage that the range of (0, 2^N-1) increases exponentially each time a specific station loses a retry attempt after collision, resulting in a higher probability that the station will lose during the next transmission attempt. Thus, a new station that has data to transmit has a higher probability of winning a collision mediation than the station having a greater number of attempts. This effect is known as the capture effect, wherein one station winning the collision mediation effectively prevents the losing station from accessing the network until the maximum number of attempts are reached. The capture effect increases substantially as more stations are added to the network. Moreover, the usable bandwidth of the network is reduced due to the increased collisions and idle time generated during the collision mediation. The capture effect also affects the throughput of a server in a client-server architecture of an Ethernet network, where clients request data from the server. The data requests from the clients typically a formed from a small number of packets, whereas the server sends substantially larger data files to the clients in multiple data packets. If network traffic causes an increase in collision rate, the throughput of the server decreases as the rate of client requests increase beause the server needs to contend equally with the clients for access to the network.

Disclosure of the Invention

There is a need for a method of accessing media of an Ethernet network that increases the throughput of the Ethernet network under heavy traffic loads. There is also a need for a method of accessing media of an Ethernet network that enables a server to maintain sufficient throughput to accommodate an increased rate of client requests.

There is also a need for mediating collisions on media of an Ethernet network that increases the efficiency of the network under heavy loads.

There is also a need for a method of mediating collisions of media of an Ethernet network that avoids subsequent collisions. These and other needs are attained by the present invention, whereby the delay time between sensing activity on the media and attempting access of the media is modified by a randomized time interval generated in accordance with a propagation delay between two network stations. According to one aspect of the present invention, a method of accessing media of an Ethernet network includes the steps of sensing deassertion of a receive carrier on the media, determining a delay time including at least a predetermined interpacket gap interval and a randomized time interval generated in accordance with a propagation delay between two network stations, and attempting access of the media in response to the sensed deassertion and after the determined delay time. The addition of a randomized time interval on the basis of propagation delay between two network stations to the delay time enables a winning station to have time to transmit data, such that other stations are able to sense the the transmitted data activity on the media. By adding the randomized time interval to client stations in a client-server arrangement, the method of the present invention enables the server to have a higher priority than the clients, minimizing collisions between the server and clients contending for the network media. The randomized time interval also provides a collision avoidance mechanism, since the increased randomness based on propagation delay enables stations to detect activity generated by a winning station.

In another aspect of the present invention, a method of managing access to a media of an Ethernet network having a server and clients includes the steps of assigning a first delay time to the server equal to a predetermined interpacket gap interval, enabling the server to access the media in response to deassertion of a receive carrier on the media and after the first delay time, assigning a second delay time to each of the clients equal to the predetermined interpacket gap interval plus a time interval randomly selected with respect to a propagation delay between two network stations, and enabling each of the clients to access the media in response to the deassertion of the receive carrier and after the second delay time. The server has a higher priority to transmit than the clients, improving throughput of the network under heavy loads. The higher priority assigned to the server also enables the server to send relatively large data files to the clients in a more efficient manner as the rate of client requests increase. Still another aspect of the present invention provides a method of mediating collisions on media of an Ethernet network, comprising the steps of sensing a collision on the media, determining a collision delay interval by multiplying a propagation delay interval representing a maximum propagation delay between two stations on the network with a randomly selected integer from a predetermined range of integers to obtain a randomized delay interval, adding the randomized delay interval to a predetermined interpacket gap interval, and then attempting access of the media in response to the detected collision and after the collision delay interval. Each mediating station randomly selects a randomized delay interval equal to at least the maximum propagation delay between two stations, minimizing occurrences of the capture effect. Moreover, mediating stations will detect the winning station by waiting a sufficient time for the transmitting signals from the winning station to propagate to the other mediating stations. Hence, the collision mediation of the present invention increases the throughput and usable bandwidth on the network, and decreases the delay per transmitted packet.

Additional objects, advantages and novel features of the invention will be set forth in part in the description which follows, and in part will become apparent to those skilled in the art upon examination of the following or may be learned by practice of the invention. The objects and advantages of the invention may be realized and attained by means of the instrumentalities and combinations particularly pointed out in the appended claims.

Brief Description of Drawings Reference is made to the attached drawings, wherein elements having the same reference numeral designations represent like elements throughout and wherein:

Figure 1 is a block diagram of a network interface according to an embodiment of the present invention.

Figure 2 is a diagram illustrating an Ethernet network.

Figures 3A, 3B and 3C are flow diagrams of the method for accessing media of the Ethernet network according to the present invention.

Figure 4 is a block diagram of the media access control (MAC) of Figure 1.

Figures 5A, 5B and 5C are graphs illustrating the improvement in network throughput by assigning a priority delay time to the server according to the present invention.

Figures 6A and 6B are graphs illustrating the network performance improvement by mediating collisions according to the present invention.

Best Mode for Carrying out the Invention Figure 1 is a block diagram of an exemplary network interface 10 that accesses the media of the Ethernet network according to an embodiment of the present invention. The accessing and collision mediation by the network interface 10 results in increased throughput on the network during heavy traffic, and a reduction in collisions.

The network interface 10, preferably a single-chip, 32-bit Ethernet controller, provides an interface between a local bus 12 of a computer, for example, a peripheral component interconnect (PCI) local bus, and an Ethernet- based network bus 14. An exemplary network interface is the Am79C970 PCnet™-PCI Single-Chip Ethernet Controller for PCI Local Bus from Advanced Micro Devices, Inc., Sunnyvale, California, and is disclosed on pages 1-868 to 1-1033 of the AMD Ethernet/IEEE 802.3 Family 1994 World Network Data Book/Handbook, the disclosure of which is incorporated in its entirety by reference.

The interface 10 includes a PCI bus interface unit 16, a direct memory access (DMA) buffer management unit 18, and a network interface portion 20 including a media access control (MAC) core 22, an attachment unit interface (AUI) 24, and a twisted-pair transceiver media attachment unit (10BASE-T MAU) 26. The AUI port 24 preferably follows the specification ISO 8802-3 (IEEE-ANSI 802.3). The interface 10 also includes a microwire EEPROM interface 28, a receive first in first out (FIFO) buffer 30, a transmit FIFO buffer 32, and a FIFO controller 34.

The PCI bus interface unit 16, compliant with the PCI local bus specification (revision 2.1), receives data packets from a host computer's CPU via the PCI bus 12. Each data packet received from the PCI bus 12 includes a header including length information identifying the number of bytes in the packet. The PCI bus interface unit 16, under the control of the DMA buffer management unit 18, receives DMA and burst transfers from the CPU via the PCI bus 12. The data packets received from the PCI bus interface unit 16 are passed on a byte-by-byte basis to the transmit FIFO 32.

The buffer management unit 18 manages the reception of the data by the PCI bus interface unit 16 and retrieves information from header bytes that are transmitted at the beginning of transmissions from the CPU via the PCI bus 12. The header information identifying the byte length of the received packet is passed to the FIFO control 34.

The Manchester encoder and attachment unit interface (AUI) 24 include a Collision In (CI) differential input pair that signals to the network interface 10 when a collision has been detected on the network media. Specifically, a collision occurs when the CI inputs are driven with a 10 MHz pattern of sufficient amplitude and pulse width meets the ISO8802-3 (IEEE/ANSI 802.3) standards. CI operates at pseudo ECL levels. The data out (DO) output pair transmits Manchester encoded data at pseudo ECL levels onto the network media 14. Similarly, the twisted pair interface 26 includes 10BASE-T port differential receivers (RXD) and 10BASE-T port differential drivers (TXD) . The media access control (MAC) 20 performs the CSM8/CD functions in response to signals from the interface 24 or 26. For example, activity on both twisted pair signals RXD and TXD constitutes a collision. Similarly, the AUI 24 detects a collision by the CI inputs. Carrier sense is detected by the Dl and RXD signal paths of the AUY port 24 and the MAU 26, respectively.

Figure 2A is a flow diagram of a method of accessing media of an Ethernet network by a client in a client-server architecture. According to the present invention, the delay time for transmitting after sensing deassertion of a receive carrier on the media is modified by adding a randomized time interval to the interpacket gap (IPG) , enabling a server to have priority on the network.

Figure 3 is a diagram of an Ethernet network implementing the media access techniques of the present invention. The network 40 includes a server 42 connected to a plurality of client stations 44. The network 40 also includes a plurality of repeaters 46 and a wiring hub 48 that provides a central connection point for the cables 50 that connect the stations 42 and 44 to the network 40. The cables 50 may be either coaxial, fiber optic, or twisted pair wire, and hence the network 40 may implement 10BASE-T, 10BASE-2, 100BASE-TX, 100BASE-T4, or 100BASE-FX networks. Thus, the disclosed network 40 may operate at either 10 megabits per second or 100 megabits per second.

The network 40 typically operates by the server 42 responding to requests from any of the clients 44. hile the debtor requests from the clients 44 tend to be small packets, the server 42 generally sends large data files, transmitted as multiple data packets across the network 40. The throughput of the network 40 is increased by prioritizing the server 42 to have a minimal delay time consisting of a predetermined interpacket gap interval when accessing the media 50. Each of the clients are assigned a delay time that includes the IPG interval randomly selected between the range of 0 and τ, whereby the upperbound τ is no more than twice the cable delay between the server 42 and the hub 48.

The access management can be embedded in each of the network interface cards of the clients 44 and the server 42. Alternately, the priority to the server 42 may be provided only when the server 42 determines that heavy loads are present on the network. In such a case, the server 42 would output a message to all the clients 44 to begin execution of the media access management of the present invention. If desired, the server 42 may include the value τ in the message to the clients 44. Typically, the value τ is equal to 2.56 microseconds on a 100 megabit per second hub.

Figures 2A and 2C are flow diagrams of the method of accessing the media 50 by the client 44 and the server 42, respectively. The process for both the client 44 and the server 42 starts in step 52 by checking whether the network interface 10 has data in its transmit fifo 32 to be transmitted on the network 40 in step 54. If the station does not have data in the transmit fifo 32 for transmission on the network, the station remains silent. If the station has data to send, the media access controller (MAC) checks in step 56 if a carrier is sensed on the media 50, indicating that the network 40 is busy. If the receive carrier is sensed, the MAC 20 waits until deassertion of the receive carrier on the media. Once the MAC 22 senses deassertion of the receive carrier on the media in step 56, the MAC 22 in each client 44 randomly selects a randomized time interval T„ between the values of 0 and τ in step 58, and then proceeds to start a delay timer in step 60a. The server 42, however, has priority over the clients

44, and has a delay time consisting of the predetermined IPG interval. Hence, the MAC 22 of the server 42 begins the counter in step 60b without randomly selecting another time interval.

The server 42 and the client 44 check the counter values in steps 62a and 62b, respectively, and increment the counters in step 64 until the delay time is reached. As shown in Figure 2C, the server 42 starts to transmit in step 66 when the counter equals the IPG value. According to the disclosed embodiment, the IPG equals 9.6 microseconds for a 10 megabit per second network, and 0.96 microseconds for a 100 megabit per second network. The delay timer in the client 44, however, continues to count for the additional randomized time interval T_R. Once the delay time reaches the delay time of the IPG interval plus the randomized time interval T_R, the MAC 22 of the client 44 proceeds to step 68 to sense if a carrier is present. If a carrier is present, then the MAC 22 of the client 44 returns to the wait state in step 56.

Thus, the addition of the randomized time interval in the delay time of the servers 44 enable the server 42 to maintain priority transmission on the network. Thus, when both server and clients are contending for access, the server will be able to transmit first at least fifty percent of the time. As shown in Figure 5, the read access tape machine 40 starts at step 60 by initializing all internal flip-flops and counters to 0. The read access tape machine 40 enters a first idle state by determining in step 62 whether a load pcm signal (ld_pcm) having a logic value 1 has been received from the audio serial output controller.

If in step 68 the client 44 does not detect a carrier, the MAC 22 and the client 44 starts to transmit in step 70. After transmission has begun in step 70, the client station 44 then checks in step 72 to determine whether a collision has been detected. If no collision is detected, then transmission is completed in step 74. If a collision has occurred with another network station, then the media access controller 22 performs a collision mediation, as shown in Figure 2B.

As shown in Figure 2B, after the collision has been detected on the media, an internal collision counter is initialized in step 76 and incremented in step 78. The collision mediating performed by the MAC 22 then checks in step 80 if the number of collisions is greater than or equal to 10. The MAC 22 then randomly selects an integer from a predetermined range in step 82, for example the range 1, 2, 3. If the number of collisions N is greater than or equal to 10, the MAC 22 then checks in step 84 if the number of collisions is equal to 16. If the number of collisions is equal to 16, then the delay time T_G is set equal to the predetermined IPG value plus the randomly selected integer multiple of the propagation delay between the two stations. The value ΔX is defined as the maximum propagation delay between any two stations, i.e., the maximum time needed between any two stations to know that there is activity on the network. The value ΔX can be up to 2.56 microseconds for a 100 megabyte per second network.

If the number of attempts is not equal to 16 in step 84, then an exponential number of access attempts is determined in step 88. Specifically, if the number of access attempts is less than 10, then the exponential number of access attempts is equal to b=2ⁿ - 1, whereas if N is greater than or equal to 10, then the value b=2¹⁰ - 1, or 1,023. After calculating the exponential number of access attempts, b, the MAC 22 randomly selects an integer value c in step 90 from the range between 0 and the exponential number b having a maximum value of 1,023. The first randomly-selected number (a) and the secondly randomly- selected number (c) are then applied in step 92 to generate the delay time according to the formula T_D=IPG +aΔX + ct_B. Thus, the value aΔX represents a randomized integer multiple of the maximum time needed between any two stations to detect activity on the network. ct„ represents a randomized time slot, whereby the slot time t_s is 51.2 microseconds in a 10 megabit per second network, and 5.12 microseconds in a 100 megabit per second network. Thus, the delay time t_d is calculated in 92 based on a two-dimensional randomized selection of the time slot t_s in a propagation delay ΔX. The combined delay time thus avoids collision by adding the X random time sufficient to enable a station to detect transmission from a winning station. In addition, the random selection from the predetermined range reduces the capture effect in stations having repeated collisions.

After the delay time is calculated in either step 86 or 92, the MAC 22 checks in step 94 whether a carrier is sensed on the media. If a receive carrier is asserted on the media, the MAC 22 waits in step 94 until deassertion of the receive carrier is sensed. If the carrier is not sensed in step 96, then the MAC 22 attempts access of the media in response to the detected collision and after the collision delay interval t_d in step 98. If a collision occurs in step 100, then the subroutine returns to step 78. Otherwise, if no collision is detected, then the procedure returns to complete transmission in step 74.

Figure 4 is a block diagram illustrating the functional components of the media access control 22. Specifically, the media access control 22 includes a plurality of registers 102, a counter 104 storing the number of access attempts (N) , a delay timer 106 that counts the delay interval, a controller 108 and a carrier sense multiple access/collision detection portion 110. As described above, the values stored in registers 102a, 102b, 102d and 102e may either be pre-loaded in an ee prompt, or downloaded from the network, for example, during a message from the server when traffic is heavy. The register 102c may store a flag indicating calculation to be performed of t_r, or may store the actual calculated value. The MAC controller 108 calculates the delay time t_d in accordance with the method shown in Figures 2A, 2B and 2C, and stores the resulting delay time in register 102F. The MAC controller 108 starts the delay timer 106 in response to a signal from the csmδ/cdllO indicating that deassertion of the receive carrier on the media has been sensed. The MAC controller 108 sends an instruction to the Manchester encoder/decoder 24 (MENDEC) to attempt access to the media after the delay timer has reached the determined delay time t_d stored in register 102F.

Claims

C AIMS

1. A method of accessing media of an Ethernet network, comprising: sensing deassertion of a receive carrier on the media; determining a delay time including at least a predetermined interpacket gap interval and a randomized time interval generated in accordance with a propagation delay between two network stations; and attempting access of the media in response to the sensed deassertion and after the determined delay time.

2. The method of claim 1, wherein the determining step comprises generating the randomized time interval having a value between zero and the propagation delay, the two network stations including a network server and a network hub, the propagation delay identifying twice a cable delay between the network server and the network hub.

3. The method of claim 2, wherein the network has a data rate of 100 megabits per second and the propagation delay equals 2.56 microseconds.

4. The method of claim 2, wherein the network further includes a plurality of clients, the determining step comprising calculating said delay time in each of the clients.

5. The method of claim 4, further comprising: generating in the server a second delay time consisting of the predetermined interpacket gap interval; attempting access of the media by the server in response to the sensed deassertion and after the second delay time.

6. The method of claim 1, wherein: the network includes a server and a plurality of clients, the determining step comprising calculating said delay time in each of the clients attempting access to the media; and the method further comprises attempting access of the media by the server in response to the sensed deassertion and after the predetermined interpacket gap interval.

7. The method of claim 1, wherein further comprising starting a delay timer in response to the sensing step, the attempting step occurring after the delay timer reaches the delay time.

8. The method of claim 1, further comprising sensing a collision on the media, the determining step comprising multiplying the propagation delay by a first integer randomly selected from a predetermined range of integers to obtain said randomized time interval.

9. The method of claim 8, the determining step further comprising: calculating a slot time interval by multiplying a predetermined slot time by a second integer randomly selected from a second range of integers calculated from an exponential number of access attempts; and adding the calculated slot time interval to the multiplied propagation delay to obtain the delay time.

10. The method of claim 9, wherein the network has a data rate of 10 megabits per second and the predetermined slot time equals 51.2 microseconds.

11. The method of claim 9, wherein the network has a data rate of 100 megabits per second and the predetermined slot time equals 5.12 microseconds.

12. The method of claim 1, further comprising storing the propagation delay in a memory of a network interface card.

13. The method of claim 12, further comprising receiving the propagation delay from the network for storage in the network interface card.

14. A method of managing access to a media of an Ethernet network having a server and clients, comprising: assigning a first delay time to the server equal to a predetermined interpacket gap interval; enabling the server to access the media in response to deassertion of a receive carrier on the media and after the first delay time; assigning a second delay time to each of the clients equal to the predetermined interpacket gap interval plus a time interval randomly selected with respect to a propagation delay between two network stations; and enabling each of the clients to access the media in response to the deassertion of said receive carrier and after the second delay time.

15. The method of claim 14, wherein the network has a data rate of one of 10 megabits per second and 100 megabits per second and said predetermined interpacket gap interval is one of 9.6 microseconds and 0.96 microseconds, respectively.

16. The method of claim 14, wherein the second delay time assigning step comprises randomly selecting the time interval from a range between zero up to and including a maximum propagation delay equal to twice said propagation delay between two network stations.

17. The method of claim 16, wherein said randomly selecting step comprises determining a cable delay between the server and a hub of said network to obtain said propagation delay between two network stations.

18. A method of mediating collisions on media of an Ethernet network, comprising: sensing a collision on the media; determining a collision delay interval comprising: (1) multiplying a propagation delay interval representing a maximum propagation delay between two stations on the network with a randomly selected integer from a predetermined range of integers to obtain a randomized delay interval; and (2) adding the randomized delay interval to a predetermined interpacket gap interval; and attempting access of the media in response to the detected collision and after the collision delay interval.

19. The method of claim 18, wherein the determining step further comprises: calculating a slot time interval by multiplying a predetermined slot time by a second integer randomly selected from a second range of integers calculated from an exponential number of access attempts by a network station; and adding the calculated slot time interval to the randomized delay interval and the predetermined interpacket gap interval to obtain the collision delay interval.

20. The method of claim 18, wherein the maximum propagation delay is approximately 2.56 microseconds.

21. A method of mediating collisions on media of an Ethernet network, comprising: sensing a collision on the media; determining a collision delay interval comprising: (1) randomly selecting a first integer from a predetermined range of integers,

(2) randomly selecting a second integer from a second range of integers calculated from an exponential number of access attempts by a network station, (3) multiplying a propagation delay interval representing a maximum propagation delay between two stations on the network with the first integer to obtain a first randomized delay interval;

(4) multiplying the second integer with a predetermined slot time to obtain a second randomized delay interval,

(5) adding the first and second randomized delay intervals to a predetermined interpacket gap interval; and attempting access of the media in response to the detected collision and after the collision delay interval.

22. The method of claim 21, wherein the second range of integers has a maximum value of 2^ID-1.

23. A network interface for connection with media of an Ethernet network, comprising: means for sensing deassertion of a receive carrier on the media; means for calculating a delay time, comprising an adder adding a randomized delay interval to a predetermined interpacket gap interval to obtain said delay time, the randomized delay interval randomly selected between zero and a maximum value determined by a propagation delay between two network stations; and means for attempting access of the media in response to the sensed deassertion of the media and after said delay time.

24. The network interface of claim 23, further comprising means for sensing a collision on the media, the calculating means comprising means for randomly selecting an integer multiple of said maximum value from a predetermined range of multiples, the adder adding the integer multiple of said maximum value to said predetermined interpacket gap interval to obtain a retry delay time, the attempting access means reattempting access of the media in response to the sensed collision and after said retry delay time.