wk5.papers: Evaluation 4

------=_NextPart_000_000C_01BE9101.FDC20BA0
Content-Type: text/plain;
charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable

The paper presents a comparison between FIFO queueing and FQ (Fair =
Queueing) algorithms and advocates FQ. The author suggests that FIFO =
queueing has some disadvantages:
* Ill-manered nodes can ignore congestion signal and consume higher =
troughput at the cost of others
* Even in the absence of ill-manered nodes, congestion control is still =
hard because it depends congestion control mechanism implemented at =
globally distributed nodes.
* Allocation of bandwith, promptness and buffer space are completely =
determinted by packet arrival at queue

The fair queueing algorithm presented is an improved version of the =
original FQ by Nagle. In this version, packets from different nodes have =
separate router's queues. The router then services the queues in round =
robin fashion. The author points out that in the original FQ, the round =
robin service implemented did not take into account packet length, =
therefore nodes with long packets consume more than their share of =
bandwith. The improved algorithm take into account the packet length in =
the round robin service, therefore is more fair.

To compare the effectiveness of the FQ algorithm and its FCFS =
counterpart, the author ran some simulations combining FQ/FCFS and =
various flow algorithms. The simulations shows that:
* Well-behaved nodes are protected from ill-manered ones, which =
encourages nodes to implement better algorithms
* The algorithm was able to deliver low delay to sources using less than =
fair share of their bandwith
* FQ presented was able to allocate bandwith, promptness and buffer =
space separately.

One main concern with the FQ presented is whether the extra logic that =
implement the fair queueing can be made fast enough in the current =
hardware. Another would be whether the router hardware can provides =
enough resources (queues and overhead) that are practical for nodes =
traffic. One main advantage of FCFS is that routers are cheap and =
simple, therefore it can handle current traffic from nodes. Author's =
simulations of network with a handful of nodes hardly convince me that =
the technique will scale well to our current needs.

------=_NextPart_000_000C_01BE9101.FDC20BA0
Content-Type: text/html;
charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable

The paper presents a comparison = between FIFO=20 queueing and FQ (Fair Queueing) algorithms and advocates FQ. The author = suggests=20 that FIFO queueing has some disadvantages:
* Ill-manered nodes can = ignore=20 congestion signal and consume higher troughput at the cost of = others
* Even=20 in the absence of ill-manered nodes, congestion control is still hard = because it=20 depends congestion control mechanism implemented at globally distributed = nodes.
* Allocation of bandwith, promptness and buffer space are = completely=20 determinted by packet arrival at queue

The fair queueing algorithm = presented is an=20 improved version of the original FQ by Nagle. In this version, packets = from=20 different nodes have separate router's queues. The router then services = the=20 queues in round robin fashion. The author points out that in the = original FQ,=20 the round robin service implemented did not take into account packet = length,=20 therefore nodes with long packets consume more than their share of = bandwith. The=20 improved algorithm take into account the packet length in the round = robin=20 service, therefore is more fair.

To compare the effectiveness of the = FQ algorithm=20 and its FCFS counterpart, the author ran some simulations combining = FQ/FCFS and=20 various flow algorithms. The simulations shows that:
* Well-behaved = nodes are=20 protected from ill-manered ones, which encourages nodes to implement = better=20 algorithms
* The algorithm was able to deliver low delay to sources = using=20 less than fair share of their bandwith
* FQ presented was able to = allocate=20 bandwith, promptness and buffer space separately.

One main concern with the FQ = presented is=20 whether the extra logic that implement the fair queueing can be made = fast enough=20 in the current hardware. Another would be whether the router hardware = can=20 provides enough resources (queues and overhead) that are practical for = nodes=20 traffic. One main advantage of FCFS is that routers are cheap and = simple,=20 therefore it can handle current traffic from nodes. Author's simulations = of=20 network with a handful of nodes hardly convince me that the technique = will scale=20 well to our current needs.

Evaluation 4 - Hadi Bandrio