Dissertation subject? - ?
?
Relationship between IT and social development?
Chose and read one article from the University of Liverpool e?-?Library that relates to your?? dissertation subject. Evaluate its content and quality.
In your answer carry out a critique of the article using the two sets of questions listed in the lecture just above the assignment section.
Adaptive Prioritization scheme for Per-
Flow Queuing based Bandwidth
Management Systems
International Symposium on Broadband Access Technologies in Metropolitan Area Networks 2006,
Oct. 5 - 6th 2006
Daniel Scheibli, and Basem Shihada
Presenter: Basem Shihada
Outline
• Motivation
• Approach
• Adaptive per-flow queuing management system (APFQ)
- Queuing Management
- Scheduler Structure
• Performance evaluation
• Conclusions
Motivation
• End users are increasingly becoming content providers on their own:
Users upload pictures to blogs or web based photo communities, operate web servers from home, use Voice over IP telephony or follow other upstream activities.
• Selected application layer protocols become higher level transport protocols - showing different traffic patterns.
For instance HTTP (e.g. attachment uploads) or SSH (e.g. tunneling)
• Even within TCP sessions, the traffic pattern might change throughout the lifetime
Approach
• Session level mediation
• Each session is given a priority
• Take the traffic characteristics into account and uses that information to dynamically adjust the priority
• Ensure a fair queuing only among all sessions that share the same priority
• Fully utilize the available network bandwidth
APFQ Queue Management
• Like in SFQ, the system creates an array with a predefined number of FIFO queues. These queues are named primary queue,
• The hashing function has been extended to consider the 5-tuple information of outbound packets. This increases the chance that
each flow can be assigned to a dedicated primary queue.
• Each primary queue has a double linked list of extend queues that handles collisions.
APFQ Scheduler Structure
• The APFQ scheduler defines different priorities
• Each priority is represented in the priority array, similar to an operating system scheduler.
• Within a priority, the active queues are represented by a double-linked chain of priority list elements
Performance Evaluation
• Defined a basic LAN with client machines connected via 100 Mbps to a gateway.
• The gateway is routing traffic between the LAN and the Internet via an asymmetric uplink with down- and upstream capacity of 768 Kbps
and 128 Kbps respectively.
• The servers are attached to the Internet via a 100 Mbps links.
• All links use FIFO based queues except the gateway's upstream link which contains the queuing scheme under test.
• Interactive bursty traffic vs. bulk data transfer
• The results are very similar showing only a marginal effect on the bandwidth
• The overall average for the FTP session is 941 and 943 ms respectively.
• For the Telnet session, the SFQ scheme produces an overall average of 986 ms; this comes from the fact that the packets of both sessions are placed in the same FIFO queue as both flows have the same 5-tuple values.
• The APFQ scheme account for an average delay of 30 ms. It keeps the FTP and Telnet flow in separate queues and ensures that the flow(s) with the highest priority is/are served next.
FTP sessions are running with time shifting.
• The first session starts at t=0, the second at t=15. After 30 seconds, the first session is stopped leaving all the bandwidth to the second session.
• APFQ favors the new arriving flows by assigning the highest priority.
• The APFQ scheduler is effective in cutting back this privilege once the bulk data transfer starts, leading to a situation where both flows are well balanced.
• The SFQ shows an oscillating effect that corresponds with the bandwidth chart data
• The APFQ privately treating the initiation of the newly arriving flow is temporarily impacting the delay times of the already existing FTP session.
• APFQ offers a potential approach for the adoption and assignment of resources by providing end users with a transparent piece of technology.
• APFQ can improve the scheduling of interactive burst oriented workloads and cut queuing delays significantly.
• APFQ is capable of dynamically adjust and change the scheduling without the need to have prior knowledge about the traffic demands, shapes, or bandwidth requirements.