Information Technology Reference
In-Depth Information
sweeping the index
n
. If five requests of the set
q
have not yet been made, its
K
-distance is thought to be infinite.
The LRU-K algorithm has, at instant
t
, a list of request data set in an
increasing order according to the
K
-distance value. In this manner it is
possible to have a deeper knowledge of which are data that are effectively
used by the server (i.e. they are requested), considering the elapsed time since
the last
K
-requests of the generic set
q
have been requested [5].
When a request data has not been found, it is searched on the disk, and
then it is made available to all users and is loaded into the cache. Because of
the limited size of the cache, it is probable that a substitution of data is
required. The data that just leave the cache is the one that has the highest
K
-
distance. It is obvious that if the parameter
K
grows, a more detailed
description of the history of the references at instant
t
is available. It may
appear that high values of
K
are associated to an improvement in the
performance of the server's response to the ubiquitous user nodes. In reality,
there is to consider that the increase of K implies the growth of the running
costs of the algorithm, making no sense an implementation of the algorithm
to exchange data in ubiquitous environment. The designers generally suggest
a value of
K
= 2 as the best compromise between quality and price in the
solution of most problems of information exchange [5].
2.8
Considerations on the use of a finite speed transmission channel
The speed of the transmission channel connecting the user node to the server
node plays a key role on the performance of information exchange. The basic
element to be considered, if the channel has finite speed of propagation is that
the data sent to the transmitter of the server node are not sent immediately
to clients who requested them, because of the 'slowness' of the transmission
channel. In this manner, queues waiting to be sent to users are formed at the
input of the transmitter. These queues slow down the server's response and
affect its global behaviour.
Three different values of transmission rate for channel were considered:
2, 155 and 620Mbps. From the analysis of the results obtained with low
values of transmission rate of the channel (e.g. 2Mbps), it has been observed
that the speed of transmission of the channel element is the bottleneck of
the server node. The behaviour of FLUSH, supported by LF-LRUs, with
different sizes of cache, has been observed and it has been noted that the
tendency to decrease the average response time as the cache grows is not
valid because of significant delays introduced by channel of transmission: in
practice, taking data from the disk or from the cache involves almost no
difference, since the delays introduced by the channel transmission override
those possibly introduced by an excessive use of the disk.
