Cells In Frames
ATM with Variable Length Packets
Dr. Lawrence G. RobertsMarch 1997
Why Cells In Frames (CIF)
CIF - ATM Protocol, but with Variable Length Packets
Cells In Frames (CIF) is ATM with variable length packets on the lines and trunks. The CIF Alliance has specified a protocol which allows ATM to be embedded into various frame based legacy protocols (Ethernet & Token Ring), using only one ATM header for up to 31 cells from the same virtual circuit in a packet. The specification of CIF over PPP and Sonet is underway. A significant feature of CIF is that ATM can be transported to workstations without changing the legacy NIC card because the necessary processing is done in simple downloaded software "SHIM" on the workstation.
The Impact of Variable Length Packets on ATM
ATM has been well engineered to support the voice, video and data traffic of tomorrow. Most importantly of all, the ATM Forum has developed an extremely effective suite of protocols for ATM which make it possible to mix voice, video and data at very high speeds. However, for various historical reasons, more relevant in the 1980ís, ATM was specified to use short fixed length cells to carry all traffic. The use of variable length packets has several important benefits over 53 byte cells:
- Lowers overhead. For typical Internet data the basic ATM overhead is 25%-30%. Variable length CIF packets can reduce this to 5%. This would be a major cost reduction on international trunks.
- Eliminates need for new Network Interface Cards (NICís) to attach workstations and servers. By permitting variable length Ethernet, Token Ring, and PPP packets to be used with all the features of ATM, CIF basically cuts in half the cost of workstation access by allowing the use of currently installed NICís.
- Eliminates the need for Segmentation and Reassembly (SAR) Hardware. This time consuming and expensive function can be avoided if variable length packets are used. Thus, CIF interfaces and NICís are inherently faster and less expensive than ATM interfaces where SARís are required.
The Original Benefits of ATM and the effect of CIF
- Fixed cell size to permit high speed switching in hardware. This was a benefit for the past decade, but today Ethernet switches have proved that with todayís silicon this is unnecessary for fast, inexpensive switching.
- Small cell size to reduce the delay for 64 Kilobit voice to 6 milliseconds for echo control. With variable length packets however, this capability is preserved.
- Small, fixed cell size has also been rumored to be important in reducing the delay of all traffic, but this has never been an issue for high-speed lines. CIF limits packet size to 1500 bytes like Ethernet and the delay variance which a full packet can cause other traffic is only one millisecond on a 10 Mbps Ethernet wire and is far less on OC-3 or OC-12 ATM trunks (voice and video need delay variance under 10-30 milliseconds). Thus, this is a non-issue for voice or video.
- Virtual Circuit (VC) Switching was chosen over Datagram switching, so as to permit fast switching speeds, enable call based QOS specification, and to enable call based flow control. The introduction of Tag Switching to IP is clear recognition that for high speed switching VCís or tags are critical. The requirement of VCís for QOS and flow control is only starting to be recognized. CIF keeps VCís.
- Quality Of Service (QOS) signaling protocol so as to permit mixing of voice, video and data on the same wire with no degradation of the delay sensitive traffic. This is an important benefit available in ATM Forum SIG 4.0 not available in IP networks today. CIF maintains the full QOS of ATM.
- QOS based routing of traffic over network trunks to enable load balancing on trunks and to ensure all traffic is routed on paths capable of supporting the QOS and bandwidth. This is another important benefit available in ATM Forum PNNI 1.0 not available in IP networks today. CIF supports PNNI.
- Low delay flow control was always recognized as a necessity for high speed switched networks. In order to control the data flow at greatly increased speeds, a much lower delay flow control, explicit rate, has been specified in ATM Forum TM 4.0, a capability not available in IP networks today. This not only reduces the buffer requirements and thus the cost of the switches but also dramatically decreases the time required for page accesses like WWW requests. CIF allows explicit rate flow control to be extended to the desktop economically over Ethernet or PPP dial-in.
Why use ATM Protocol?
The ATM Forum finished a complete 4.0 protocol suite for ATM in April 1996. ATM switches without 4.0 have really not had any significant advantage over todayís LANs except speed. However, as ATM switches come out in 1997 with the complete 4.0 protocol they will be able to support mixed voice, data and video very effectively over both the LAN and the WAN. The elements of 4.0 are as follows:
- UNI Signaling 4.0 - Allows signaling of bandwidth and delay requirements for QOS.
- TM 4.0 - Specifies Explicit Rate flow control and QOS functions.
- PNNI 1.0 - Specifies QOS based inter-switch routing.
No other protocol suite in history has existed which included signaled QOS, Explicit Rate flow control, or QOS routing. All are critical to operating a network with low delay and QOS. However, this protocol suite need not be locked to ATM switch hardware, it could be used on any switch platform and over any link protocol. CIF is the use of the full ATM protocol stack but with variable length packets on the lines.
Explicit Rate Flow Control
One of the critical improvements which ATM and CIF have, which is not available in TCP/IP today, is Explicit Rate Flow Control. Flow control is used to control the traffic sources so that they do not send too much data into the network at any moment. If a trunk in the network is overloading, all the sources using that link must be told to slow down. This is absolutely necessary for a data network because the self-similar characteristics of data traffic are such that simply under-loading the network will not work. The critical difference between flow control techniques is the time it takes to tell the source about congestion and to get it under control. Today, TCP is typically used in LANs to control the data flow. TCP was created 15 years ago when networks were much slower. It operates a binary rate flow control algorithm between the two end-stations, slowing down the data flow if the return path indicates that data was lost. If data is not being lost, it speeds up the flow. It always oscillates with a period of 1-2 seconds on the Internet, losing data each cycle and under-utilizing the network on the other part of the cycle. Its characteristic time is about one second, that is the time it takes TCP to stop the data flow due to increased congestion.
Figure 1. Comparision of TCP and Explicit Rate Time to Control
The ATM Forum has now completely specified a new generation of flow control, explicit rate flow control, which operates about 200 times faster than binary rate flow control protocols like TCP. Explicit rate flow control operates with small control messages called Resource Management (RM) cells being sent around the network to convey the flow control information back to the source. This way, a higher priority path can be used for the RM cells so that they do not need to be delayed by the data. In CIF they consume 1% of the network capacity. When a switch sees a RM cell it marks it with the highest data rate it can currently support for the associated source. This way, when congestion occurs, the switch can quickly mark RM cells with a new rate which will eliminate the congestion and these cells can move at the speed of light back to the sources telling them exactly how fast to send data. Compared to todayís TCP the delay comparison is:
Cause for Speedup for Explicit Rate over TCP
- Priority for RM cells over data. - Typically in Internet today a factor of 10
- Explicit rate rather than oscillation - Typically oscillation is another factor of 5
- Switch to source distance vs round trip - Typically a factor of 2-4
Thus, explicit rate can stop the source at least 100 times faster than TCP
Benefits of Explicit Rate Flow Control
Because explicit rate controls traffic at the source 100 times faster than TCP in Wide Area Networks (WANís), the data which must be stored in the network switch buffers is also reduced by the same factor. IP routers typically are configured with enough memory to support data flow from all inputs for 1 second, whereas ATM switches are only configured with about 10 milliseconds of memory, 100 times less than the TCP based routers. This reduction is made possible by explicit rate flow control. This major reduction in control time has several major benefits (in the WAN environment):
- Explicit rate (ATM and CIF) switches require 100 times less memory than IP Routers or Ethernet switches, thus reducing the cost of the switches by at least a factor of two to seven.
- Explicit rate switches have 100 times less delay variance the IP switches, down from 1 sec. to 10 ms. This dramatically improves the utility of the data service for interactive activities.
- Startup transmission rate on an explicit rate network can be 100 time faster than on a TCP controlled network because the switches need the same size buffer to absorb 100 times the rate for 1/100 the time. This means that World Wide Web (WWW) page access would be 100 times faster than it is today, down from 10 seconds to 0.1 seconds. This improvement requires adjusting or avoiding TCPís slow-start when explicit rate connections are complete end-to-end.
- When explicit rate is complete end-to-end, there is virtually no data loss (10-12) whereas TCP/IP has 10-20% data loss (and retransmission) in peak hours today on the Internet. Thus, 20% would be saved on line cost.
A major problem for International Carriers with ATM is the extremely high overhead ATM has for the typical 220-byte Internet packet. Not only is the ATM cell header 10%, but the last cell of an AAL-5 packet wastes half a cell or an additional 12%. On top of that, most protocols like classical IP carry the IP header in each AAL-5 packet adding another 9%. IP has an overhead of 9% plus an additional 20% retransmission of packets due to TCP packet loss at peak hour. Thus, IP and ATM (with explicit rate) have about the same 30% overhead. CIF with explicit rate, however, only has a 5% overhead, a 25% saving over either ATM or IP!
Figure 2. Overhead for ATM, IP and CIF on Sonet Trunks
Cells In Frames Design Overview
The CIF specification1, 2 defines the protocol to use across an Ethernet or Token Ring connection between a CIF workstation and a CIF switch. Since a standard Ethernet or Token Ring NIC is used, there is software "SHIM" loaded into the workstation between the NDIS and the driver. The "SHIM" adds the CIF header to packets before they are transmitted and removes the header when they are received. It queues outgoing data into multiple queues for QOS management. It also processes the RM cells for explicit rate control. Except for the addition on the "SHIM" a CIF workstation is exactly the same as an ATM workstation. It has ATM signaling software and runs both native ATM applications as well as standard IP applications. Figure 3 shows the workstation, switch and the additional, downloaded software.
Figure 3. CIF workstation and switch
CIF Edge Switch
CIF edge switches differ from simple Ethernet switches in that:
- They interpret the CIF header and ATM VP/VC in addition to other protocols.
- They support QOS with multiple queues and weighted fair queuing to manage the delay on these queues. This addition would also required if RSVP were to be used to support QOS.
- They mark the Available Bit Rate (ABR) Resource Management (RM) cells when they are congested with the rate that they could maximally support on this VC (explicit rate marking).
- The software supports ATM signaling.
The first three of these changes require changes to the ASIC chips used for switching but do not significantly change its cost. The fourth item (software) does require an additional 8 Megabytes of memory and thus adds a small increment of cost to the overall switch, but only a few percent. Thus, there is every reason to expect CIF edge switches to cost about the same as Ethernet switches.
CIF provides a way to obtain the full functionality of ATM flow control and QOS over legacy frame based media. In addition to being less expensive than ATM for workstation access due to the elimination of the need for a new NIC, its use of variable length packets results in a line overhead 25% less than ATM or IP.
Pros and Cons of Various Switches in the LAN
Capability IP Routers ATM Switching CIF Switching QOS Support Yes-RSVP Yes-SIG 4.0 Yes-SIG 4.0 QOS Routing No Yes-PNNI 1.0 Yes-PNNI 1.0 ER Flow Control No Yes-TM 4.0 Yes-TM 4.0 Use existing NIC Yes No Yes Cost to Desktop Low High Low Trunk Overhead 30% 30% 5%
The importance of supporting ATM protocol across the final workstation link is that QOS and explicit rate flow control must be implemented end-to-end in order to be effective. If the final link is too expensive and users opt to use simple Ethernet switching with no QOS or flow control, then the benefit of ATM Protocol in the backbone is lost. However, with CIF Edge switches, the added cost over Ethernet switching is likely to be so small that the benefits of QOS and flow control far outweigh any cost difference. In the WAN backbone, the line efficiency of CIF is compelling over ATM or IP and CIF has all the other benefits of ATM. Thus, there is good reason to consider utilizing CIF throughout the whole network.
- The CIF specification can be found at http://cif.cornell.edu
- Roberts, Lawrence G., ATM Forum 96-1104, Request for Coordination of Cells In Frames Specification, August 19, 1996
Copyright © 2001 Dr. Lawrence G. Roberts
Some material from: The ATM Handbook 1997 published by McGraw-Hill.
Reprinted with permission.