Вы находитесь на странице: 1из 10

Comparative Analysis of Performance Proxy Server

Using Algorithms Greedy Dual Size Frequency


(GDSF) and Least Recently Used (LRU) Network
Internet at Pondok Pesantren Yasin Kudus
Ivandi Julatha Putra, Ari Sriyanto Nugroho, Sri Anggraeni Kadiran
Department of Electrical Engineering
Polytechnic State Semarang
Semarang, Indonesia
ivandijulatha @ yahoo.com, ari.sriyanto@polines.ac.id, rinikadiran67@yahoo.co.id

Abstract- In realizing pengingkatan Internet access needs of I. BACKGROUND


the Pondok Pesantren Yasin Kudus facilitated network
infrastructure. The network infrastructure is equipped with a
In today's modern era of information and communication
proxy server that also serves as a proxy cache. Algorithms cache technology are necessary, especially in the internet service. At
removal is necessary as a method to remove the cache the boarding school, students are young people who in the
automatically to keep the cache still has room for new objects in future will color the life of the nation. As one component of
the cache. There are various types of algorithms deletion deletion the younger generation, students also have to and master the
while each algorithm has the properties of each. Therefore we latest technological developments, including information
need the proper selection and deletion algorithms in accordance technology and communications.
with the requirements of access to the Pondok Pesantren Yasin In the field of information and communication
Kudus network. This final project contains a comparative
technologies, today Pesantren YASIN have subscribed to
analysis algorithms Greedy Dual Size Frequency (GDSF) and
algorithms based on Least Recently Used (LRU) based on the
Internet access at an Internet Service Provider (ISP) with a
parameters of the hit ratio, byte hit ratio, response time and size bandwidth of 3 Mbps. With the number of students currently
distribution. Tool used SquidAnalyzer, Calamaris and SARG, about 216 people, which is not too large bandwidth is sorely
the value of the average hit ratio LRU with 19.22% scenario, the lacking when done accessing the rush hour, the afternoon after
average hit ratio GDSF by 27.28%, while the average hit ratio is school until the afternoon before Maghrib time. This is what
taken directly LRU 16.53 %, the average hit ratio directly GDSF makes the Internet feels slow because by using only a small
5.77%, the average byte hit ratio LRU with a scenario of 30.99%, bandwidth but the number of clients who are very much the
the average byte hit ratio is the scenario GDSF 40.30%, the very high load.
average byte hit ratio LRU is taken directly by 13.93%, the
average byte hit ratio GDSF directly 10.10%, obtained an
average response time of 4,075,491 LRU scenario millisec, the
In order to optimize the network infrastructure of the
average value of response time in the scenario GDSF 6010. 538 Internet is built, then use a proxy server as bridging the
millisec, the average value of the response time is taken directly internal network to the Internet network lodge. By utilizing a
LRU 197 088 588 millisec, the average value of response time proxy cache server that can serve to save the Internet data
directly GDSF 182 147 206 millisec. Size distribution LRU through it. In time these data are requested again by the same
showed a tendency to save objects by size and popularity. While client or another client have its on the proxy server. Proxy
the size distribution showed a tendency to maximize GDSF small server does not need to retrieve data from a web server source
object that is popular and minimize the large object that popular. but simply take the data from the cache that have been stored
By looking at the ratio of the value that exists when the maximum
on the proxy server itself.
object enlarged from 10 MB to 500 MB performance of algorimta
GDSF the better Based on a comparative analysis algorithm
cache removal to the scenario as well as directly when students However, if the existing cache storage space on the proxy
use the Internet access, as well as considering the activity of the server has a capacity that is fixed so there will be storage on
client in the network Pondok Pesantren Yasin Kudus, the LRU the proxy cache has been filled up with objects. It is thus
algorithm chosen as the best cache removal algorithms ..
necessary the removal of objects in the cache or so-called
Keywords- Proxy Server;Cache Replacement; LRU;GDSF; cache replacement policy [1]. In this case there are several
kinds of algorithms such as Greedy Dual Size Frequency
(GDSF) and Least Recently Used (LRU).

1
Proxy Server acts as a gateway to access the Internet for proxy server. With scenarios in pengakesannya it will look
each computer on the network. Proxy servers are not seen by perfomansi of the two algorithms are. 5 Laptops used as a
the client computer. A user who interacts with the Internet via medium for data capture and each laptop to access the website
10 different. Here is a list of web diguanakan table for the
a proxy server will not be aware that a proxy server is
scenario.
handling requests accomplishments [2].
Table 1. List web scenario used
Previously been many studies that discuss the proxy PC1 PC2 PC3 PC4 PC5
server, one of which is the implementation of the proxy server www.detik www.polin http://wiki.sq http : http://edition.
.com es.ac.id uid- //www.liputa cnn.com/
as a cache replacement policy by comparing the best algorithm (accessed (accessed cache.org/ n6.com/ (accessed 1x)
based on testing of algorithms GDSF or algorithms LFUDA 10x) 6x) (accessed 2x) (accessed 7x)
on the network intenet wholesale markets Setono pekalogan http://www http://www. http://www.q http://www.tri http://squidan
the measured parameter hit ratio, byte hit ratio , response .kompas.co lazada.co id q.com/ bunnews.com alyzer.darold.
m/ / (accessed (accessed 1x) / (accessed net/ (accessed
time, and size distribution [1]. Even so, with a network of (accessed 5x) 8x) 6x)
wholesale markets implemented on pekalongan Setono are 2x)
other determining factors such as the test conditions, the http : http://indox http://www.su http://www.b http://www.bi
effects of proxy server configuration and using that disk space //www.viv xi.net/ ara.com/ ola.com / ng.com/
for the cache. a.co.id/ (accessed (accessed 7x) (accessed 7x) (accessed 1x)
(accessed 7x)
7x)
Other journals that discuss the cache replacement http://www http://www. http: / http://snapfile http://wallpap
algorithm to perform testing of 3 algorithm is GDSF, LFUDA .pulsk.com imdb.com/ /www.goal.co s.com/ ercave.com/
/ (accessed (accessed m/id-ID/ (accessed 1x) (accessed 5x)
andLRU. Then to conclude the algorithm which has a 1x) 3x) (accessed 4x)
maximum performance, the required variables or parameters http://dikti. http://filehi http://www.b http://www.o http:
of research as a method of comparison is shown through the go.id/ ( ppo.com/ bc.com/ kezone.com/ //www.wowk
analysis of squid are Cache Cache manager and Calamaris. accessible (accessed (accessed 5x) (accessed 9x) eren .com /
8x) 10x) (accessed 2x)
Comparative parameters used to evaluate the performance of http://www Http://www http://www.el http://www.to http: // www
the cache replacement algorithm that is Request Hit Ratio, .beritasatu. .metrotvne evenia.co.id/ urnamentsoft .livescore.co
Byte Hit Ratio, and Median Service Time [3]. However, in com/ ws.com/ (accessed 2x) ware.com/ m / (accessed
this journal a very little number of users so that research (accessed (accessed (accessed 3x) 3x)
1x ) 1x)
results have a narrow scope, and not enough data to prove the
http://jiathi http://www. http://www.c http : http://www.g
effectiveness of the use of GDSF as the best algorithm. s.com/ sciencedire hange.org/ //www.ubaya. unadarma.ac.i
(accessed ct.com/ (accessed 4x) ac.id/ d/ (accessed
In this study, we propose the performance of Proxy Server 2x) (accessed (accessed 1x) 5x)
3x)
is implemented on the Internet network Pondok Pesantren
http://www http: // http://www.id http://www.la http: // www.
Yasin Kudus using GDSF algorithm or LRU algorithm as a .indonesia.t www.fujits .undp.org/ mudi.co.id/ bhinneka.com
cache removal. The purpose of this study was to determine the ravel/en u.com/id/ (accessed 1x) (accessed 1x) / (accessed
performance of the proxy server in serving our clients and see (accessed (accessed 1x)
the effectiveness of a cache removal using one of the 5x) 3x)
http://www http://www. http://www.jp http://www.pi http://www.so
algorithms to be applied to the net Pondok Pesantren Yasin .indonetwo indiatimes. nn.com/ kiran- lopos.com/
Kudus. Performance parameters used in this study is the hit rk.co.id/ com/ (accessed 9x) rakyat.com/ (accessed 5x)
ratio, byte hit ratio, response time, and size distribution. (accessed (accessed (accessed 3x)
4x) 8x)
http://www http: // http://harimul http://warungf http://www.ja
II. METHODS .sukadi.net/ www.yosbe ya.com/ iksi.net/ goanhosting.c
(accessed da.com/ (accessed 1x) (accessed 1x) om/ (accessed
A. Testing Methods 10x) (accessed 1x)
2x)
Testingagainst two algorithms deletion cache (cache
replacement algorithm) that LRU algorithm and GDSF. Tests
on the algorithm does this by accessing the web in a scenario Furthermore, the testing was conducted at the time
that is already in the list and then on the run algorithm one by students requesting access the internet and then recorded in the
one each week. The first test uses LRU algorithm for two HTTP object of squid access.log. From access.log will be
days, then after GDSF algorithm was used for two days with a processed by software Squidanalyzer, Calamaris, and SARG
hit ratio parameters, byte hit ratio, response time and size are displayed in the form of a web report. Cache memory is
distribution. Performed at 08:00 to 09:00 hours. used for 1024 MB. Furthermore, the stored maximum object
size and a minimum size of 100 MB size 64B. Then, the
During the two days of each algorithm every request from parameters used are Hit Ratio, Byte Hit Ratio, Response Time,
the client where web access is already in the list beforehand and Size Distribution. Data collection was performed within
any website that will be accessible to so will be served by a 24 hours. With GDSF Algorithm for 3 days and LRU

2
algorithm for 3 days. Access client that the workload will be pass on to the ISP to request the object. With the number of
very helpful in testing the performance of the two algorithms clients who are dipondok also affect the performance of the
are. Different client activity also provides testing of the cache from squid because if a lot of the same object often
performance of each algorithm .. accessed it will also speed to be accessed and will be
prioritized for removal because GDSF algorithm works by
B. System object to the size of an object that is stored is likely to be
The system will be builtaims to provide Internet access to removed. Meanwhile, when the client is requesting a different
the boarding school. Client in the cottage through a device object the LRU algorithm will prioritize the elimination of the
connected to the network access point spread in the cottage. cache by looking at the object that is stored in the old cache is
The access point is then connected to the office via the not requested by the client will be prioritized for removal.
antenna point to point. In the office, there is a router device, So if there is a client that accesses the same object, Squid
server and modem GPON. Routers in the boarding school into will continue the object that has been stored in the cache. This
a device that connects network and server client access to the can save time in accessing client because the requested object
Internet through a modem GPON. So that the network was in is located in a local network.
boarding school can connect to the Internet ..
C. Testing Parameters
Hitratio is the most popular parameter for measuring the
efficiency of the cache replacement algorithm. Hit ratio is a
measure of comparison of the number of all objects cached by
the number of requests sent by the client to cache [3]. So you
can use the following formula:

[1]

The second parameter is the byte hit ratio. Byte hit ratio is
the ratio of byte values are cached with the total bytes
obtained client. Byte hit ratio indicates the percentage of
network bandwidth savings made. Byte hit ratio of 30%
indicates that three out of the 10 bytes requested by the client
retrieved from the cache. The seven bytes else was taken from
the original server [3]. Use the following equation

[2]
Figure 1. Design of Network System Response time is a parameter that indicates the total
elapsed travel time client to access the object. Values are
In Figure 1 is described on a general overview of network
calculated from the initial response time client performs a
topology on the system. PC Server is used as a proxy server
request object to obtain the object. Response time value is
and also PCs as monitorung network traffic monitoring. Server
inversely proportional to the hit ratio and byte hit ratio. If the
that will be built using Debian Linux operating system which
object is in the cache by the client, the response time will be
is then installed packages and a proxy for traffic monitoring
faster than taking an object from the original server. However,
software used cacti. In the squid proxy server is used which is
there is no guarantee that if the hit ratio and byte hit ratio is
installed in the server package. Squid proxy server to function
high, the response time will be too low. little objects with high
determine the performance Least Recently Used (LRU) and
downloading latency may have a lower response time
Greedy Dual Size Frequency (GDSF) as a cache removal
compared with many objects in the cache with lower latency
algorithm.
downloads [4].
By using 2 ISP load balancing router gives a chance to
share the traffic load. G-Media to ISP subscription with 3
Mbps bandwidth while Biznet subscribe to the 25 Mbps
bandwidth. Router will also transfer the web from the Internet
[3]
access to the proxy first client. Thus, all access is done by a
Size distribution is a parameter that describes the
proxy server. Squid proxy server stores the object that is taken
deployment of objects in the cache based on size. With size
directly from the Internet into the cache. Then, if in the cache
of the requested object has not been saved then it will directly distribution, will be known performance of an algorithm in
storing objects according to size. Size distribution was able to

3
explain in details the hit ratio of byte values are obtained. algorithm. This caused a data erasure method GDSF prioritize
Within the parameters of size distribution can be shown the large object to be removed. So that a large object size is more
percentage of requests to specific object size. likely to be removed to make more space available to store
many small objects that could potentially increase the hit ratio.
D. Applications used Testing
GDSF algorithm has a high hit ratio value. With such an
Squid Analyzer is one tools belonging Squid log analyzer increase in hit ratio on the second day. In addition there are
that can provide complete statistics covering a time, hits, factors that have a website accessible to the small size makes
bytes, users, network, until the top url and top domain most GDSF better performance in this scenario. LRU algorithm is
frequently accessed. The advantages offered from Squid less likely to hit ratio value for accessing the first day and
Analyzer them is no need for any database to be run, and also second day are treated equally. This is due to the nature of the
see Squid This analyzer includes a fairly user friendly [5]. LRU see the object of the request based on the last time in a
Squid Analysis Report Generator (SARG) is one long time accessible prioritized for removal. On the second
application that can help generate Internet usage statistics. day LRU and GDSF have increased compared to the first day
SARG generates web-based reports. SARG is used to retrieve because of objects in the previous request is stored in the
the data response time or travel time client of the request cache makes the hit ratio increases. The average yield LRU
object to obtain [6]. algorithm produced the hit ratio of 19.22% and 27.28% GDSF
Calamaris are additional applications that can analyze the algorithm. From these results look GDSF scenario using the
access logs of squid and generate reports on the performance algorithm has better performance.
of the use squid as a proxy server. Calamaris installed on the
proxy server and produces web-based reports. Calamaris used In the experimental results Hit Ratio measurements
to retrieve data about the hit ratio, byte hit ratio, as well as the directly when students access the Internet by two algorithms
size distribution [7]. for 3 days for each graph comparison algorithm can be
described as follows.
III. RESULTS
Quality of service refers to the ability of service on the
network traffic, to determine the quality of a network, network
Hit Ratio The Scenario
services can be a better level, planned and controlled. Quality 25 23.11
of service parameters measured include delay, jitter, packet
Persentasi (%)

20
loss, and throughput.Testing value for each quality of service
parameters obtained from the average - average per test, the 15 11.05 11.33
10 7.89
Null SAP with four hosts in VPLS 1000. While, in SAP dot1q 5.64 LRU
3.13
is as much as 8 hosts each are on VPLS 5800 and 9000 with 5
GDSF
each service has a client as much as 4 host.Laptops are used in 0
testing has different specifications - different, and has 1 2 3
aEthernet 100 Mbpssupport.
Hari
A. Hit Ratio
In experimental measurement results hit ratio with a Figure 3. Results are directly hit ratio when students access the internet
scenario with two algorithms each day at 8:00 a.m. to 09:00 2
can be described as the following comparison chart. In Figure 3 it can be seen that within three days of testing,
the LRU algorithm has a higher hit ratio on all days of testing
against GDSF. This makes the algorithm performs better hit
ratio on the Internet at boarding Yasin. It is influenced of total
requests made by different clients in three days of testing on
the first day total request from the LRU 89 239 by 9946 while
the total number of hits request of GDSF 42 973 by 2597.
Accessing the total hit a different client greatly affect the
performance of the proxy server on each each algorithm. On
the second day LRU Hit Ratio reached 23.11% as many
objects that have been stored in the cache is large or small due
to the nature of the object size LRU memperahatikan not but
notice based on the time when the object is accessed apabula
Figure 2. Results of hit ratio is the scenario the longer time it is accessed also be potentially deleted
priority. For GDSF algorithm on the second day of a very
in Figure 2 can be seen with the testing for 2 days with a small value of 3.13% hit ratio with the case due to different
time of 1 hour at 8 a.m. to 09:00 in the scenario using 5 object accessed on daily visits. The third day of the LRU
Laptop as a client to each client 10 websites. GDSF algorithm remains efficiently by 11.3% compared with GDSF that has
is shown to have a higher hit ratio is compared with the LRU hit ratio of 7.89% although the total request from the LRU on

4
the third day amounted to 48 398 or less than the total request
GDSF to 58856. On the third day of the LRU can effesien
because many objects that have been accessed cacheable make
Byte Hit Ratio Directly
every client that accesses back proxy does not need to take on 30 26.73

Persentasi (%)
the Internet. From table 4.6 average yield direct hit ratio LRU 18.13
20 15.13
algorithm is 16.53% and 5.77% GDSF algorithm. It is seen 12.23
that the LRU algorithm is better. 10 5.98 5.42 LRU
B. Byte Hit Ratio 0 GDSF
On the results of experimental measurement scenario byte 1 2 3
hit ratio is accessing the Internet with two each algorithm Hari
within 2 days at 8:00 a.m. to 9:00 o'clock comparison chart
can be described as follows. Figure 5. Byte Hit Ratio Test Results directly when students access the
Internet
Byte Hit Ratio Scenario Figure 5 shows the performance of the two algorithms on
60.00 53.55 byte hit ratio. LRU algorithm is able to obtain initial byte hit
Byte Hit Ratio %

38.84 ratio is higher than with GDSF. But when the second day of
40.00 27.01 testing, the byte value GDSF higher hit ratio. Then on the third
19.59 day of testing byte highest hit ratio remains GDSF.
20.00 LRU
This happens because the LRU save the object without
0.00 GDSF regard to size, which means that despite the large object size
1 2 will still be cached. This can be seen in the table: Percentage
distribution LRU byte object hit by a category VII on a
Hari Thursday, July 27th (the first day of testing) has a value that is
large enough percentage of 32.51% compared to a category
Figure 4. Testing Results Byte Hit Ratio scenario VII in GDSF algorithm which has a value of percentage of
Based on theFigure 4 can be seen the percentage of byte 0%. It shows that on that day LRU able to transmit large byte
hit ratio in the scenario, GDSF have byte hit ratio is high. This taken from the resulting byte cache hit ratio is high.
is due, in theory LRU cache removal method is based on the Then on the second day of testing GDSF can generate
length of time the object is in the cache. So the files located in value byte hit ratio higher than LRU. This can occur because
the cache the longest it will be prioritized for removal. While of the success of GDSF maintain a small object and popular
GDSF attention object stored size to the extent that it will be increasingly referred to by the client so that even a small
prioritized for removal. Judging Byte first day GDSF hit ratio object sizes but many referred add value byte hit ratio.
has a value of Byte hit ratio 27.01% with a scenario of Meanwhile, the performance of LRU rose on both the size
accessing the website do the same first and second day of the distribution byte object category VII hit ratio increased to
second day the higher value for the first day of the request 42.97% but GDSF more efficiently with a small object in
object has been first stored cached and on the nature of the category III with a percentage of 20.73%. On the third day of
object makes GDSF attention object having a large size is the performance LRUturun for the tests according to Table
replaced with new objects that can make a proxy performance size distribution of byte hit LRU, byte hit of object categories
can be maximized. While LRU first day with the byte value small increase (I, IV, V) whereas byte hit object large category
ratio hit 19.59 low because the same scenario of accessing (category VI and VII) decreased so that it makes the byte
websites and recurrent cause LRU cache in the removal of less value hit LRU ratio decreases. On average byte hit 13.93% for
than the maximum. Object at the same time into the cache so the LRU algorithm and average algorithm in table 4.8 GDSF
its elimination longer. Most bytes hit taken from the original 10,10%. This indicates the LRU better than GDSF based byte
server (Internet). On the second day LRU has a byte value hit hit ratio.
38.84% and GDSF has a byte value hit 53.55%. Table 4.2 of
C. Response Time
the average byte hit ratio LRU algorithm is 30.99% and
40.03% GDSF algorithm. It makes GDSF algorithm showed a In the experimental results of measurement of response
good performance. time in the scenario to access the Internet with two each
algorithm within 2 days at 8:00 a.m. to 09:00 hours chart
On the results of byte hit ratio measurement experiment comparison can be described as follows.
conducted by students directly accessing the Internet with two
algorithms for 3 days for each graph comparison algorithm
can be described as follows.

5
ratio value indicated on the first day and the second has a
Response Time Scenario value of the high byte hit ratio is supported by the nature of
the LRU based on the time length of the file in the cache.
8,000,000 Client is accessing the network more frequently accessing the
6,107,256 5,913,820
Total millisec

6,000,000 4,141,529 4,010,353 same web and repetitive that cause LRU has a high response
time. On the third day of the LRU value lower response time
4,000,000 due to the number of requests from a client that slightly
LRU
2,000,000 compared with the first and second day. So with a small
0 GDSF request and less then the number of objects too small which
1 2 makes the travel time becomes small.
While the algorithm GDSF on the first day and the second
Hari has a small response time of the day due to the many objects
taken from the proxy cache so that the travel time from the
Figure 6. Results of Measurement of Response Time is the scenario client requesting the object to the receiving object becomes
shorter. It can be seen from the value of the first byte hit ratio
in Figure 6 is shown the value of a response time of two with small object categories that make accessing the Internet,
algorithms with milisec time value. First Day LRU algorithm the client has a response that cepta. On the day ketdua with a
has a lower response time compared with GDSF algorithm. total of bytes requested from client 389.848976 able to be
This could happen because of the scenario with five client serviced by the byte cache hit reaching 104 379 314 having
access website each client 10 websites and also has been previously requested a lot of objects that are already in
determined the number of times the website is accessed / the cache. Then on the third day response time is very
requested several times. Thus during the first day LRU is more valuable because it is a new increase in the request object from
efficient for accessing the web 10 with some websites do the client. With the new request then have to take on the
access more than 1 time. For the first day GDSF less efficient original server (ISP) that requires a longer response time. And
because objects that have been determined to do repeatedly a request to the object major categories affect the value of the
accessed cache removal GDSF make less work because of the response time. That's because the numbers byte hit ratio which
nature GDSF attention to size object to be removed. Then, leads to high response time of low value so that the request
with the number of objects taken from the original server / byte hit ratio high and proxy cache relative faster filled lead-
Internet, causing GDSF require greater time to object to the time service to the client algorithm GDSF very nice on the
client. Table 4.3 of the average response time of the LRU first day of supported algorithms GDSF have the nature of
algorithm millisec 4,075,941 and 6,010,538 GDSF millisec. attention to size object where if the large size of the object will
LRU looks capable of providing a fast time of GDSF. be prioritized for removal replaced with a new object. From
the response time results in Table 4.9 and Table 4:10 average
In the response time measurement results of the of 197 088 588 millisec LRU algorithm while the average
experiment performed by the students directly when using the response time algorithm GDSF at 4:10 table at 182 147 206
Internet access for 3 days for each graph comparison millisec.
algorithm can be described as follows.
D. Size Distribution
Response Time
45926290 In size distribution measurement experiment results hit in
the scenario to access the Internet with two each algorithm
500000000 35195386 5
within 2 days at 8:00 to 09:00 hours chart comparison can be
3
Total millisec

400000000 described as follows.


300000000 15713752
200000000 2
LRU
Size Distribution Hit The First
82174380
75224646
100000000 11954066
GDSF
Day Scenario
0 60
50
Persentasi (%)

1 2 3
40
Hari

20 22.46
17.67 LRU
Figure 7. Results of Testing Response Time directly when students access the
internet
8.88 17.7419.6 11.44
14.6616
4.69 GDSF
In Figure 7 is shown Response Time value of both 0 1.56 5.08 0 0 0
algorithms directly when students access the Internet to show I II III IV V VI VII VIII IX
the time in millisec. Response time performance of the
algorithm LRU on the first day and the second can be seen in Kategori Ukuran Object
the picture response time 4:41 has a high value and on the Figure 8. Results of Measurement of Size Distribution hit in the first day
third day response time low. It happened because the byte hit scenario

6
In Figure 8 shows that the LRU algorithm is a smaller Size Distribution Byte Hit The
percentage of the algorithm GDSF hit ratio for all sizes of
objects from I to VI, this is because the LRU have a tendency
First Day Scenario
to insert objects into the cache without notice the size of the 60
object. On the other hand, maintains GDSF category III and 55
IV. Then the object with category V also inserted into the 40
cache but the number is almost the same between LRU
18.6620.4718.55
algorithm and GDSF. Means that the algorithm minimizes 20 13.66 LRU
GDSF object with category V for entry into the cache. This is 16.6716.67
10.09 GDSF
evidenced by the value of their masinng higher percentage 0 0 1.547.34 0
0 0
GDSF category. I II III IV V VI VII VIII IX
Kategori Ukuran Object
Size Distribution Hit
Secara The Second Day Gambar 10. Measurement Results Size Distribution byte hit in
the first day scenario
Scenario
Figure 10 shows the distribution byte hit sizes of object II
40 up to the GDSF algorithm higher than the LRU algorithm
31.3
34.48 because in the first day scenario GDSF performance to
30 minimize objects with large categories has not worked it can
20.03 29.68
20 14.65 21.73 20 LRU be seen from the value in category VI that reaches 55% LRU
16.47 algorithm is only 16.67%. For categories II through IV the
10 GDSF GDSF algorithm maintains the object and maximizes it. Then
7.3 6.96 7.14
3.57 in object V the number of byte hit percentages of the LRU and
0 0 0 0 GDSF algorithms is almost identical that it shows GDSF
I II III IV V VI VII VIII IX minimizes the Category V object to enter the cache. LRU
Figure 9. Size Distribution Measurement Result hit by the second day scenario algorithm tends to insert large objects without seeing the
object size seen from Figure 4.36 in categories II to VI has
In Figure 9 shows that the size distribution at the first increased the percentage value.
object bigger than the LRU algorithm GDSF algorithm caused
by accessing the website with the number 50 on 5 client with
each client 10 has a different Characteristics between websites Size Distribution Byte Hit
make LRU value higher object in category I compared GDSF.
But in categories II, III, IV algorithm GDSF hit ratio has a The Second Day Scenario
value higher than the LRU to maintain the object. That's
because GDSF attention to size objects stored in the cache, the 50
V and VI categories tend to be maintained. This is evidenced 40 41.67
by the LRU algorithm which has a high percentage value in 29.24 32.07
categories V and VI. Because LRU tends to keep objects large 30
and cached. 22.26
20 22.72 LRU
12.38 18.78
17.55 17.85
10 GDSF
5.91
0 0 0 0 0
I II III IV V VI VII VIII IX
Kategori Ukuran Object

Gambar 11. Measurement Results Size Distribution byte hit by the second day
scenario

Figure 11 shows that the objects II through IV of the


GDSF algorithm are higher byte hit ratio than the LRU
algorithm. This is because GDSF maintains small objects to
maximize. In the Category V and VI objects the LRU
algorithm is higher byte hit ratio than the GDSF algorithm

7
because LRU maintains a large-size object to be inserted into and the GDSF only 389,848,976 was the factor with the small
the cache regardless of object size. request will affect the percentage of each category of object so
small percentage. With a small number of bytes GDSF has a
In the result of measurement size distribution test done by higher byte hit ratio than LRU. For category VI GDSF has hit
santri directly when using Internet access for 3 days for each ratio of only 1.95% and LRU hit ratio of 12.86% it is also
algorithm can be described graph comparison as follows. influenced by the number of clients who request the category
of object becomes less visible performance than GDSF. In
category VII LRU has high hit ratio with 48% percentage
Size Distribution Hit Ratio which means able to give byte of big object category to client.
For GDSF still consider the size of the object to be stored in
Directly Day One the cache by minimizing the category of large object that is in
60 category VII with the percentage of 11.11%.
56.88
Persentasi (%)

40 31.6
LRU
20 16.73 24.32 Size Distribution Hit Ratio
12.32 GDSF
13.28 6.59
1.433.13 11.83 Directly Day Third
0 0.870.71 0 0 0 0
I II III IV V VI VII VIII IX 40
37.51

Persentasi (%)
Gambar 12. Measurement Results Size Distribution hit directly when santri 30 27.89 25.66
Internet access the first day 23.62
20 17.11
In Figure 12 it shows that the categories of objects I LRU
10 2.97 8.08 3.23 12.05
through IV LRU algorithms have higher hit ratio percentages 1.36 GDSF
9.36 10
than the GDSF algorithm this is due to requests from different 0 1.18 2.85 0 0
clients on the LRU and GDSF algorithms. In the V VDSF
I II III IV V VI VII VIII IX
object category has a high value due to many requests from
clients on the object compared to LRU. While in the category Kategori Ukuran Object
VI and VII GDSF tend not to defend the object because of the
nature of GDSF that keeps the size of the object to be stored in Gambar 14.Measurement Results Size Distribution hit directly
the cache is different from the LRU which maintains the when santri third day Internet access
category of object VI with the hit value ratio of 6.59% and
category VII has a value of 24, 32% tend to retain the large Figure 14 shows that objects I, III, IV, V, VI LRU are
object is then inserted into the cache. higher than GDSFs caused by requests from different clients
on the LRU and GDSF algorithms. In category VII GDSF has
a higher hit ratio compared to LRU because on that day the
Size Distribution Hit Ratio client merequest these objects differently making GDSF more
effective increase in providing services requested by the client.
Directly Day Second
60 In the experimental results measurement size distribution
48
Persentasi (%)

41.06 byte hit that dilukaukan santri directly access the Internet with
40 28.23
20.18 two algorithms each within 2 days at 08.00-09.00 can be
20 12.86 LRU
1.135.71 11.04
23.16
19.56 11.11 described graph comparison as follows.
0 0.162.96 1.95 0 0
GDSF
I II III IV V VI VII VIII IX Size Distribution Byte Hit Ratio
Kategori Ukuran Object Directly Day One
60 46.52 53.51
Measurement results Size Distribution hit directly
Persentasi (%)

Gambar 13.
when santri Internet access the second day 40
32.51
20 14.93 11.15 LRU
2.85
Figure 13 shows the objects I through IV LRU algorithms 0 0 14.21 10.04
11.28
0 0.67 0 0 0 0 GDSF
higher than the GDSF algorithm this is caused by the number I II III IV V VI VII VIII IX
of client requests to each different object category resulting in
a high byte hit ratio on the LRU algorithm. In the object size Kategori Ukuran Object
category V LRU decreased with the percentage of 20.18% and
GDSF 19.56% then can be seen from table 4.7 on the second Gambar 15. Measurement Results Size Distribution byte hit
day the number of bytes from the LRU reached 4,793,034,194 directly when santri Internet access the first day

8
performance in hit bytes. Means of all category III, IV, and V
Figure 15 shows that the percentage value of the category object requests are able to provide better hit bytes
distributed byte hit ratio of the LRU algorithm ratio is higher than GDSF. However for all requests category II, VI, and VII
than the GDSF algorithm in the categories of objects I, II and are served well by GDSF over LRU. Thus the request from the
III. This is because many objects in categories I, II, and III client also mempengaru byte hit value of each algorithm.
have been stored to make LRU algorithm higher size
distribution byte hit ratio in that category. While in the IV. CONCLUSION
category of V VDSF object size has a percentage of 53.51% From the results of the research, the following conclusions can
inversely proportional to the percentage of LRU algorithm be drawn:
which only 11.15% can occur because many of the same 1. The performance of the LRU algorithm and the GDSF
object is inserted into the cache. For category VI, VII and VIII
algorithm as cache deletion by testing in a scenario is as
the GDSF algorithm reaches 0% because it maximizes objects
with category V to enter the cache. With LRU having a follows:
deletion properties based on the time of the object in the ● The LRU and GDSF algorithms both maintain popular
cache, large-capacity objects tend to be retained and for small objects that are in the cache. However, the difference lies in
categories tend to be deleted. the GDSF algorithm considering object size. Large-size
objects tend not to be retained in the cache. While LRU
consider the old object that is in the cache.
Size Distribution Byte Hit Ratio ● Size Distribution of object size in LRU is likely to include
Directly Day Second popular and large objects
60 ● Size distribution of object sizes on GDSF is likely to
Persentasi (%)

42.97 maximize popular small objects and minimize popular large


40 38.39
24.220.73
17.9817.67 objects
20 LRU
0 5.7 11.85 18.51 ● Average hit ratio (total request cache can be serviced) LRU
0 0 2.83 13.46 2.94 0 0 GDSF algorithm 19.22% of the average total of all client requests
I II III IV V VI VII VIII IX 12488, and average hit ratio (total request that can be served
Kategori Ukuran Object cache) GDSF algorithm 27.28% of the average total of all
10337 client requests.
Gambar 16. Measurement Results Size Distribution byte hit ● Average Byte hit ratio (total request byte taken from the
directly when santri Internet access the second day cache) LRU algorithm is 30.99% of total byte request
159.377.564 and average byte hit ratio (total request byte
Figure 16 shows that the percentage of bytes hit the object
size categories I, II, III, VI, VII, and VII LRU show better retrieved from the cache) algorithm GDSF 40 , 30% total
performance in the hit byte. Means that of all object requests junlah byte request 219.544.923.
category I, II, III, VI, VII, and VII LRU sizes are able to ● Average Response time (LRU algorithm) is 4,075,491
provide better hit bytes than GDSF. However the request of millisec and average response time (client access time gets the
the V GDSF object category is better than LRU. requested object) GDSF 6.010.538 millisec algorithm.
● From the tested parameters it can be concluded that when
Size Distribution Byte Hit Ratio testing with GDSF algorithm scenario is more effective than
LRU algorithm.
Directyly Day Third 2. Performance of proxy server on the Internet network at
60 Yasin Kudus Pesantren when students use Internet access are
Persentasi (%)

40 39.05 as follows:
19.75 21.28
20.42 ● Size Distribution of object size in LRU is likely to include
20 5.5 LRU
0 3.32 6.69 7.08 popular and large objects
0 0 1.1 7.342.886.45 0 0 GDSF ● Size distribution of object sizes on GDSF is likely to
I II III IV V VI VII VIII IX
maximize popular small objects and minimize popular large
Kategori Ukuran Object objects
● Average Hit ratio (total request cache can be serviced) LRU
Gambar 17. Measurement Results Size Distribution byte hit
algorithm is 16.53% of the total average total client request
directly when santri third day Internet access
73234 and average hit ratio (total request that can be served
Figure 17 shows the result of the hit byte percentage that cache) 5.77% GDSF algorithm from the average of the total
for Category III, IV, and V object sizes show better client request is 52884.

9
● Byte hit ratio average (total request byte taken from the
cache) LRU algorithm 13.93% of total byte request
3.906.476.980 and average byte hit ratio (total request byte
taken from the cache) GDSF 10 algorithm, 10% of total junlah
byte request 2,219,231,276.
● Average Response time (LRU LRU algorithm 197.088.588
millisec) and average response time (Client access time gets
the requested object) GDSF algorithm 182.147.206 millisec
● LRU performance shows better than GDSF on Hit Ratio and
byte hit ratio parameters while response time parameter,
GDSF performance is better than LRU.
● Based on comparison analysis and weighing size
distribution of access client network of Yasin Kudus
Pesantren, selected LRU algorithm as the best cache removal
algorithm.

APPRECIATION
The author would like to thank Mr. Ari Sriyanto Nugroho and
Mrs. Sri Anggraeni Kadiran for his guidance during the
process of preparing the paper. Next is a thank-you to Yasin
boarding school that has allowed me to build Internet network
infrastructure and State Polytechnic Semarang as an institution
that I have been studying.
BIBLIOGRAPHY
[1] Sadewa, EE (2016). Analisis Komparasi Cache
Replachement Algorithm Untuk Meningkatkan
Kinerja Proxy Server Pasar Grosir Pekalongan.
Semarang.
[2] Yani Kris Nanang, RA (2011). Membangun Firewall
dan Proxy Server Untuk Membatasi Hak Akses
Internet di SD Negeri Piyaman II. 6.
[3] Mardi, A. (2011). Analisis Perbandingan Algoritma
Penggantian Cache Pada Squid Berdasarkan
Parameter Request Hit Ratio. 1-16.
[4] John Dilley, MA (1999). Enhancement and Validation
of Squid's Cache Replacement Policy.
http://www.hpl.hp.com/techreports/1999/HPL-1999-
69.html.
[5] Rahman, R. (2014).
http://www.sayabisa.com/2014/02/memonitor-kinerja-
squid-proxy-dengan.html. Retrieved 07 22, 2017, from
http://www.sayabisa.com/2014/02/memonitor-kinerja-
squid-proxy-dengan.html:
http://www.sayabisa.com/2014/02/memonitor-kinerja-
squid-proxy-dengan.html
[6] Rahman, R. (2013). Mahir Administrasi Server dan
Router Linux Ubuntu 12.04 LTS. Bekasi.
[7] Saini, K. (2011). Squid Proxy Server 3.1 Beginner's
Guide. Birmingham: www.packtpub.com.

10

Вам также может понравиться