Browse Prior Art Database

Method to adaptively scale performance on network media with asymmetric rates

IP.com Disclosure Number: IPCOM000010223D
Publication Date: 2002-Nov-06
Document File: 5 page(s) / 110K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method to adaptively scale performance on network media with asymmetric rates. Benefits include improved performance.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 41% of the total text.

Method to adaptively scale performance on network media with asymmetric rates

Disclosed is a method to adaptively scale performance on network media with asymmetric rates. � � Benefits include improved performance.

Background

        � � � � � An asymmetric digital subscriber line (ADSL) controller has a receive bandwidth that is much higher than the transmit path. The device driver (host software) for the controller is a connection-oriented deserialized asynchronous transfer mode (ATM) driver that does not reject packets for transmission. The drive queues all the packets handed to it by upper layers and does not have a way to signal the upper layers to stall transmits before the drive is ready to accept and send more packets again.

        � � � � � The controller utilizes queues populated by host software to transmit packets. However, the transmit queue has a limitation on the number of transmit buffers that the controller can handle at one time.

        � � � � � Telephone companies worldwide use different logic link control/standard network access protocol (LLC/SNAP) protocols to deploy ADSL on their telephone networks (see Figure 1). In the case of LLC/SNAP encapsulation, the ADSL driver stack solution exposes a connection-less 802.3 local area network (LAN) interface to the protocol layers. LLC encapsulation is required when several protocols are carried over the same virtual circuit (VC). The rest of the network stack, such as transmission control protocol/Internet protocol (TCP/IP), is unaware of the connection-oriented nature of the underlying host software driving the ADSL network device.

        � � � � � If there is a burst of upstream traffic, from the customer premises equipment (CPE) to the server, the driver can silently drop packets, which adversely affects unreliable protocol traffic, such as user datagram protocol (UDP). All of the packets sent down to the driver must be queued and accessed at a number, n, at a time where n stands for the number of transmit buffers that the controller can retain in its transmit queue. This solution improves throughput for upstream traffic. However, this affects downstream throughput when simultaneous upstream and downstream traffic occurs.

        � � � � � When the system is running only downstream traffic from the server to the CPE, the throughput is at full rate. The bulk of the transmits are TCP acknowledgements (ACKs) that are sent out to the server without much latency. However, when upstream traffic is also involved, the acknowledgements are no longer the bulk of the transmits and are interleaved with the upstream packets in the ATM layer’s queue. This change occurs even though the TCP acknowledgements are sent out with a higher priority at the TCP protocol layer. The deep queue of transmits in the ATM layer increases the latency of these acknowledgements and affects downstream throughput adversely. The latency of the TCP ACKs has an almost linear relationship with the TCP window size. A higher TCP window size usually im...