Evaluation of CPU Consuming, Memory Utilization and Time Transfering Between Virtual Machines in Network by using HTTP and FTP techniques

Document Sample
Evaluation of CPU Consuming, Memory Utilization and Time Transfering Between Virtual Machines in Network by using HTTP and FTP techniques Powered By Docstoc
					                                                    (IJCSIS) International Journal of Computer Science and Information Security,
                                                    Vol. 9, No. 6, June 2011




     Evaluation of CPU Consuming, Memory
Utilization and Time Transfering Between Virtual
 Machines in Network by using HTTP and FTP
                   techniques.

              Igli TAFA, Elinda KAJO, Elma ZANAJ, Ariana BEJLERI, Aleksandër XHUVANI

              Polytechnic University of Tirana, Information Technology Faculty
                             Computer Engineering Department
                                       Tiranë, Albania
        itafaj@gmail.com, e_kajo@yahoo.com, ezanaj@gmail.com, arianabejleri@yahoo.com,
                                                axhuvani@yahoo.com

  Abstract: In this paper we want to evaluate Transfer             Hypervisor than those machines without it. Another
  Time, Memory Utilization and CPU Consuming                       problem is Physical Memory utilization and data
  between virtual machines in Network by using FTP and             overhead during live migration phase. Some
  HTTP benchmarks. As a virtualization platform for                researchers in [13], [14] presented some methods of
  running the benchmarks we have used Xen hypervisor
                                                                   memory overbooking and compression of this
  in para-virtualization mode. The virtual machine
  technology offers some benefits such as live migration,          memory in virtual machine, in order to improve
  fault tolerance, security, resource management etc. The          memory utilization and performance of migration.
  experiments performed show that virtual machines
  above the hypervisor consume more CPU, memory and                In this paper we analyze Transfer Time, CPU
  have bigger transfer times than in a non virtualized             Consuming and Memory Utilization between Virtual
  environment.                                                     machines and physical machines by using FTP [15]
                                                                   and HTTP requests [13]. All results are presented in
  Keywords: Transfer Time, Memory Utilization, CPU                 respectively tables.
  Consuming, Virtual Machines, Xen-Hypervisor.
                                                                   This paper is organized as follows. Section II dis
     I. INTRODUCTION                                               scribes the experimental architecture. Section III
                                                                   presents the experimental evaluation. Section IV
  Virtual machine technology offers a lot of benefits as           presents conclusions and outlines areas of future
  shown             in           previous        research          work.
  [1],[2],[3],[4],[5],[6],[7],[8] such are live migration,
  fault tolerance, security, resource management and                    II. EXPERIMENTAL ARCHITECTURE
  reduced energy consumption. Some virtual machine
  technologies are based on a software layer called                The Figure 1 and Figure 2 we present the basic of
  Hypervisor [9]. There are [10] three main types of               experimental architecture. In Figure 1 there are 2
  virtualization: full virtualization, OS virtualization           computers which are connected with UTP cat 7 cable
  and para-virtualization. Para virtualization approach            using Twisted Pair technique. Communication of two
  gives more flexibility than others. Based on this                computers is Full duplex. Both computers can
  approach we can use ESX-Server [ 11] or Xen [] .                 communicate with each other by network fast-
  Because Xen is free open source and implements                   ethernet interface 100/1000 Mbit/sec.     In each
  ballooning method [12] we have used this Hypervisor              computer we have setup the virtual machine
  .                                                                environment with Xen 4.1 and CentOS 5.5 as Dom0
                                                                   operating system.
  Anyway, the virtualization technology offers some
  “black holes”, for example the Hypervisor introduces
  a slight delay during a transfers from one machine to
  another one [13]. Also CPU consumes more
  functionalities in machines which include the



                                                             100                               http://sites.google.com/site/ijcsis/
                                                                                               ISSN 1947-5500
                                                       (IJCSIS) International Journal of Computer Science and Information Security,
                                                       Vol. 9, No. 6, June 2011




                                                                      Evaluation of transfer time in FTP server

                                                                           Evaluation of transfer time between 2 VM on
                                                                            the same Host

                                                                           Evaluation of transfer time between 2 VM on
                                                                            different Hosts

                                                                           Evaluation of transfer time between 2 physical
                                                                            machines connected by twisted pair cable.

                                                                           Evaluation of transfer time between 3 Physical
Fig.1 Communication between 2 Physical Machine with Twisted                 Machines connected by Gigabit Switch
Pair. Above those hosts are Guest Virtual Machines.
                                                                      Initially we have used 2 physical machines which
In figure 2 we have installed 3 computers connected                   support 2 virtual machines, by           using para-
with Gigabit switch. The topology of routing is Bus.                  virtualization approach (XEN 4.1). In both machines
Communication is Full Duplex. We used             a                   we have installed CentOS 5.5 as Dom0. In DomU1
management Gigabit Cisco Switch, but we could use                     and DomU2 respectively we have installed Scientific
simple switch too.                                                    Linux 6.0 and Ubuntu 10.04 Server for the first
                                                                      machine. In the second machine in DomU1 is
                                                                      installed Ubuntu 10.04 Server. Initially we want to
                                                                      test the transfer time from a client to a server
                                                                      between 2 VM in the same physical host by using
                                                                      FTP technique. We will repeat the test by using 2
                                                                      virtual machines in different physical hosts and
                                                                      finally we will evaluate this time between 2 physical
                                                                      machines connected by twisted pair technique. We
                                                                      want to transfer ISO image (XP.ISO with SP2 = 557
                                                                      MB). In the machine with Scientific Linux 6.0 we run
                                                                      a FTP client and in Ubuntu 10.04 a FTP server. To
                                                                      realize the transfer of XP.ISO file from one machine
                                                                      to another we have used Samba FTP tool (which is
                                                                      part of Scientific or Ubuntu Server). We can measure
                                                                      the time of file transfer from the start moment at
                                                                      source machine to the destination machine.
Fig 2. Three Computers connected with a Gigabit Switch. Above         The experiment is repeated again with Scientific
Host computers are setup Virtual Machines
                                                                      Linux 6.0 client machine as DomU1 in host 1 and
                                                                      Ubuntu 10.04 Server as DomU1 in host 2 which is
The architecture of all the machines is X86 - 64 bit,                 used as FTP server.
RAM 4 GB. CPU Quad-Core, supported with VT
and Hyper-threading technology.                                       Finally we will evaluate the time transferred between
                                                                      2 physical machines, respectively host 1 and host 2.
     III. EXPERIMENTAL EVALUATION:                                    Host 2 will serve as FTP server and Host 1 as a FTP
                                                                      client. The results are presented in table 1:
The evaluation is separated in three phases:
                                                                        TABLE1 THE EVALUATION OF TIME TRANSFERRED OF
     Evaluation of transfer time for FTP and Web                           XP ISO IMAGE BY USING SAMBA FTP TOOL
      servers
                                                                      Time     transferred   Time      transferred   Time    transferred
     Evaluation of CPU consumption for FTP and                       between 2 VM on        between 2 VM on         between 2 Physical
      Web servers                                                     the same Host          different Hosts         Hosts
                                                                      (Host 1)
     Evaluation of memory utilization for FTP and
                                                                      48 sec                 86 sec                  36 sec
      Web serves




                                                                101                                   http://sites.google.com/site/ijcsis/
                                                                                                      ISSN 1947-5500
                                                   (IJCSIS) International Journal of Computer Science and Information Security,
                                                   Vol. 9, No. 6, June 2011




In Figure 2 we have 3 computers connected with a                  Evaluation of CPU consuming in FTP Server
Gigabit Switch. All computers communicate with
each other by Fiber channel. This communication                   We want to evaluate the CPU consuming of Web
will increase the performance of network (As it                   Server which means how much is the percentage of
knows communication with fiber channel utilize                    CPU dedicated to our experiment. It is presented in
more bandwidth and speed than UTP cable). The                     average values during the test in figure 1 and figure
third computer is a clone of second computer. Now                 2. To monitor the CPU consuming we have to used
we want to evaluate the transfer time between 2 VM                xentop command in /proc and System Monitor into
in different hosts (i.e between DomU1 in host 1 and               System Administrator Menu. Both of them offer an
DomU1 in host3). Then will transfer the image from                explicit form of CPU consuming including all located
computer 1 to computer 3. The results are presented               processes into computer by calculating the average
in table 2                                                        value [ (DomU1+DomU2+…)/n where n is the
                                                                  number of virtual machines (The same thing would
 TABLE 2 THE EVALUATION OF TIME TRANSFERRED OF                    be with physical machines) ] of these processes in
         XP ISO IMAGE BY USING SAMBA FTP
                                                                  virtual or physical machines in our experiment. To
                                                                  calculate the total CPU consuming we have built a
Time transferred between 2     Time transferred between 2
VM     on    different Hosts   Physical Hosts (Host 1 and
                                                                  script in C which gives a formula:
(DomU1 in Host 1 and Dom       Host 3)
U1 in Host3)                                                      Running process rate x nr of active process +
                                                                  Sleeping process rate x nr of sleeping process= Total
83 sec                         32 sec                             CPU Consuming                                     (1)

                                                                    TABLE.3 AVERAGE OF CPU CONSUMING DURING THE
                                                                     TRANSFERRED OF 557 MB BASED ON FIGURE 1 AND
                                                                                     FORMULA 1
Let`s analyze Tab. 1 and Tab. 2.
                                                                  CPU      consuming   CPU       consuming     CPU     consuming
In Tab. 1 the time transferred between 2 VM on the                between 2 VM on      between 2 VM on         between 2 Physical
same host is small. This is because the transfer of 557           the same Host        different Hosts         Hosts
MB image file from one VM to another one                          (Host 1)
performed over the interfaces of the same computer
architecture (In reality we are in the same computer).            61,4%                62,2 %                  55.1 %
If we compare time transferred between 2 VM on
different host as it looks from table 1, total time
transferring is 86 sec. The reasons are:
                                                                    TABLE. 4 AVERAGE OF CPU CONSUMING DURING THE
     The transferred speed of the file between 2                    TRANSFERRED OF 557 MB BASED ON FIGURE 2 AND
      computers over the network is more slowly                                       FORMULA 1
      then the internal interface of computer
      architecture (ISA interface).                               CPU consuming between 2           CPU consuming between 2
                                                                  VM     on    different Hosts      Physical Hosts (Host 1 and
     Media communication between hosts is based                  (DomU1 in Host 1 and Dom          Host 3)
      on UTP cat 7 which is more slowly then other                U1 in Host3)
      medias (i.e fiber media, which is presented in
                                                                  61,45%                            55.16%
      table 2)

Also in table 1 we show that time transferred between
different hosts is 36 sec (< 86 sec and < 48 sec). In
the both cases the main reason is the delay that is               As it look from table 3 CPU consuming between 2
introduced from the Hypervisor.                                   VM is higher than between 2 physical machines. The
                                                                  reason is a part of CPU, consumes to maintenance
In table 2, the time decrease slightly. The reason is             the Hypervisor. (61,4 % > 55,1 %). CPU consuming
media communication. In table 2 we are using                      doesn’t affect from the computer communication in
Gigabit Switch and Fiber channel communication                    network. This is the reason that in table 4, CPU
while in table 1 we are using only UTP cable.                     consuming has the same values as table 3.




                                                            102                                 http://sites.google.com/site/ijcsis/
                                                                                                ISSN 1947-5500
                                                        (IJCSIS) International Journal of Computer Science and Information Security,
                                                        Vol. 9, No. 6, June 2011




Evaluation of memory utilization in FTP Server                          Hosts (DomU1 in Host 1 and       1 and Host 3)
                                                                        Dom U1 in Host3)
We want to evaluate the physical memory utilization
                                                                        Host 1 MAX= 1,06 GB of           Host 1 MAX=687 MB of RAM
during the transferring .iso image from one virtual
                                                                        RAM                              (≈17 % of RAM)
machine to another, inside a physical host. Then we
want to evaluate the Physical memory utilization                        Host 2 MAX= 1,08 GB of           Host 2 MAX= 711 MB of
while .iso image transferring between two virtual                       RAM                              RAM    ( ≈ 17 % of RAM)
machines in different physical hosts. Initially we
have used Mem_Access [13], but now we have used                         Both (≈ 41 % of RAM)
this tool to evaluate the Mem-Utilization during file
transfer. We have implemented this tool to a script
which is write in C and called MemC. For each 10
MB transfer from Server Machine to Client Machine                       In table 5 and table 6 we presented that memory
it calculates Memory Utilization by activation this                     utilization doesn`t affect from communication in
tool. The result is resolve in a record. This record is
                                                                        networks, but it affects from presence of the
part of “My SQL” Data Base which is installed in
Server Machine (mysql Ver 12.21 Distrib 4.0.14, for                     Hypervisor (from 41 % to 17 %)
pc-linux). Average Memory Utilization is Total
Sum of record dividing with nr of records. Always                       Evaluation of memory utilization and CPU
we can use Samba FTP server for transferring data                       consuming in Web Server
from one machine to another one. Finally we repeat
the experiment by using transfer between 2 physical                     We will repeat the above experiment by using Web
hosts (fig 1). Table 5 give the results for memory                      server instead of FTP Server. Initially we will
utilization between 2 virtual machines.                                 accomplish the experiment inside a physical machine
                                                                        between 2 VM. Then we will repeat the test between
TABLE.5 PHYSICAL MEMORY UTILIZATION DURING THE                          2 Physical Machines (Fig 1).
 .ISO FILE TRANSFER BY USING SAMBA FTP BASED ON
                          FIGURE 1.                                     We have installed LAMP (Appache 2 and My SQL
                                                                        client , My SQL Server) in Web Server Virtual
Average Memory        Average Memory        Average     Memory          Machine or Web Server Physical Machine.
utilization between   Utilization between   Utilization between
2 VM on the same      2 VM on different     2 Physical Hosts            We have built another script in C++, called MemCP
Host (Host 1)         Hosts (Dom U1 in      (Host 1 is a Client         which get information from MemAccess benchmark
                      Host 1 is a Client    FTP and Host 2 is a         for every request ( From Client machine to Server
                      FTP and Dom U1        Server FTP)
                                                                        machine) and previously script which was located in
                      in Host 2 is a
                                                                        /proc. This script make a calculation by adding the
                      Server FTP)
                                                                        results got it in module of Apache 2 installed in our
MAX = 1,06 GB of      Host 1 MAX= 1,06      Host 1 MAX=687              machine. This module is implemented in order to
RAM (≈41 % of         GB of RAM             MB of RAM (≈17              include the MemAccess tool for each request. The
RAM)                                        % of RAM)                   results obtained present the Memory utilization in
                      Host 2 MAX= 1,08                                  each Virtual Machine for each process which is
                      GB of RAM             Host 2 MAX= 711             located in host computer. Total results can present in
                                            MB of RAM    (≈             percentage. Each request from Client Machine to
                      Both (≈ 41 % of       17 % of RAM)                Server Machine performed by using Httperf
                      RAM)                                              Benchmark which can generate 10 request in Second.
                                                                        The time duration of experiment is 1 minute. One
                                                                        request is equal to 10 MB which corresponds to
                                                                        .html file located in /home. Results are presented in
If we repeat the experiment based on figure 2 we will                   Table 7.
present the results in table 6.
                                                                          TABLE 7 MEMORY UTILIZATION IN WEB SERVER BY
TABLE 6 PHYSICAL MEMORY UTILIZATION DURING THE                                     USING HTTPERF BENCHMARK.
 .ISO FILE TRANSFER BY USING SAMBA FTP BASED ON
                     FIGURE 2.                                          Memory utilization     Memory Utilization   Memory Utilization
                                                                        between 2 VM on        between 2 VM on      between 2 Physical
Average Memory Utilization       Average Memory Utilization             the same Host          different   Hosts    Hosts (Host 1 is a
between 2 VM on different        between 2 Physical Hosts (Host                                (Dom U1 in Host 1    Client and Host 2 is




                                                                  103                                http://sites.google.com/site/ijcsis/
                                                                                                     ISSN 1947-5500
                                                        (IJCSIS) International Journal of Computer Science and Information Security,
                                                        Vol. 9, No. 6, June 2011




(Host 1)            is a Client and Dom   Web Server)                  TAB.9 TIME REQUEST            IN WEB SERVER BY USING
                    U1 in Host 2 is                                    HTTPERF BENCHMARK.
                    Web Server)
                                                                       Time request in      Time request in         Time request in
(≈37 % of RAM)      Web Server (≈37,7     Web Server (≈16,6            Web        Server    Web           Server    Web        Server
                    % of RAM)             % of RAM)                    between 2 VM on      between 2 VM on         between 2 Physical
                                                                       the same Host        different Hosts         Hosts
                                                                       (Host 1)

                                                                       38 sec               66 sec                  33 sec
   TAB. 8 CPU CONSUMING IN WEB SERVER BY USING
                HTTPERF BENCHMARK.

CPU     consuming   CPU      consuming    CPU     consuming
between 2 VM on     between 2 VM on       between 2 Physical           If we make a comparison between Tab.9 and Tab.1,
the same Host       different Hosts in    Hosts   in   Web             the time request for Web Server is smaller than Time
(Host 1) in Web     Web Server            Server                       transferred in FTP (38 sec < 48 sec, 66 sec < 86 sec).
Server
                                                                       This time is not affect so much in third case. In this
                                                                       case we are not using Hypervisor (33 sec < 36 sec)
57,4%               57,5 %                53,6 %

                                                                       The reasons are :

                                                                            Request in Httperf are smaller, so that doesn’t
If we compare Table 7 and Table 5, Memory                                    give any effect in degradation of CPU.
Utilization during File (557 MB) transferring in FTP
Server is bigger than Memory Utilizes in Http                               Memory utilization of Httperf ( Tab.7) is
Request in Web Server (41 % > 37 %). The reasons                             smaller than Memory Utilization of FTP
are:                                                                         (which uses Samba FTP tool)
     FTP uses 2 connections in communication                               IV. CONCLUSIONS
      Client/Server. One for synchronization and
      one for data transmission, while Web Server                           FTP approach consumes more CPU, Time and
      uses only 1 connection. Each request reserves                          Memory Utilization than Web Approach
      a small amount of memory.
                                                                            The Hypervisor offers an additive time for
     Introduction of the Hypervisor in Web Server                           both approaches, but it is significantly in FTP
      during the requests from Client to Server                              approach
      offers a smaller effect than FTP Server
                                                                            Transfer Process between 2 Physical
     If the computers don`t use Hypervisor during                           Machines offer a better performance than
      transfers,    Memory       Utilization     is                          between 2 Virtual Machines.
      approximately the same.
                                                                            Introduction of good environment decrease the
If we compare Table 8 and Table 3, CPU consuming                             time consuming between machines and
during file transfer in FTP Server is bigger than Http                       increase the CPU performance. (Fiber channel
request in Web Server. The reasons are the same as                           is better environment than UTP channel)
Memory Utilizes.
                                                                       In the future we want to test the performance and
Evaluation of Time Transferring in Web Server.                         CPU and memory utilization by using:

Finally, we can repeat the experiment by using                              Live Migration between 2 machines
HTTPerf Benchmark. We want to evaluate the time
duration after 56 requests from client machine to                           Memory Compaction Algorithms and their
Web Server machine. Every request is 10 MB. So we                            effects in Memory Utilization
will test approximately in the same conditions the
results of table 9 with table 1 (10x56 requests =560                        Memory Ballooning approach and it`s effect
MB in Web Request ≈ 557 MB in FTP)                                           in CPU consuming.

                                                                            Extension of these experiments in WAN.



                                                                 104                                 http://sites.google.com/site/ijcsis/
                                                                                                     ISSN 1947-5500
                                                         (IJCSIS) International Journal of Computer Science and Information Security,
                                                         Vol. 9, No. 6, June 2011




References:                                                              [11] Carl Waldspurger, 2006, “Memory Resource Management in
                                                                         VMWare ESX Server”
                                                                         [12] Weiming Zhao, Zhenling Wang, 2009, “Dyanamic Memory
[1] Hien Nguyen Van, Fr´ed´eric Dang Tran, 2009, “Autonomic
                                                                         Balancing for Virtual Machines”.
virtual resource management for service hosting platforms”
                                                                         [13] Jin Heo, Xiaoyun Zhu, Pradeep Padala, Zhikui Wang,2010,
[2] Michael Cardosa, Madhukar R. Korupolu, Aameek Singh,
                                                                         “Memory Overbooking and Dynamic Control of Xen Virtual
2007, “Shares and Utilities based Power Consolidation in
                                                                         Machines in Consolidated Environments”
Virtualized Server Environments”
                                                                         [14] Hai Jin, Li Deng, Song Wu, “Live Virtual Machine Migration
[3] Keller, G.;       Lutfiyya, H.; Dept. of Comput,,2010,
                                                                         with Adaptive Memory Compaction”
“Replication and Migration as Resource Management Mechanisms
                                                                         [15] Moreira, Miguel Elias M. Campista, Lu´ıs Henrique M. K.
for Virtualized Environments”
                                                                         Costa, and Otto Carlos M. B. Duarte, 2007, “OpenFlow and Xen-
[4] Ando, R.; Zong-Hua Zhang; Kadobayashi, Y.; Shinoda,
                                                                         Based Virtual Network Migration”
Y.; ,2009,” A Dynamic Protection System of Web Server in Virtual
Cluster Using Live Migration”
                                                                         Authors Profile
[5] Takahiro Hirofuchi Hirotaka Ogawa Hidemoto Nakada Satoshi
                                                                         Igli TAFA. He is a pedagogue in Polytechnic University, in
Itoh Satoshi Sekiguchi ,2009, “A Live Storage Migration
                                                                         Computer Engineering Department. In 2008 he has finished
Mechanism over WAN for Relocatable Virtual Machine Services
                                                                         the Master Thesis and now is PhD student. His PhD topic
on Clouds”
                                                                         according to Virtual Machines direction.
[6] Moghaddam, F.F.; Cheriet, M.; 2010, “Decreasing live virtual
                                                                         Elinda KAJO (MECE). She is a pedagogue in Polytechnic
machine migration down-time using a memory page selection
                                                                         University, in Computer Engineering Department. She has
based on memory”
                                                                         finished the PhD thesis at 2004 in Object Oriented
[7] Wei Wang; Ya Zhang; Ben Lin; Xiaoxin Wu; Kai
                                                                         Programing direction.
Miao; ,2010, “Secured and reliable VM migration in personal
                                                                         Elma ZANAJ. She is a pedagogue in Polytechnic University
cloud”
                                                                         in Computer Engineering Department. She has finished the
[8] Anton Beloglazov* and Rajkumar Buyya, 2010, “Energy
                                                                         PhD thesis at 2009 in Computer Sensor Network direction in
Efficient Resource Management in Virtualized Cloud Data
                                                                         Polytechnic University of Ancona Italy.
Centers”
                                                                         Aleksandër XHUVANI. He is a chief of Computer Software
[9] Andrew Tanenbaum, 2009, “Modern Operating System”
                                                                         Department in Polytechnic University. He has finished the
[10] Espen Braastad, 2006, “Management of high availability
                                                                         PhD study at Bordeaux in France. At 2004 he is graduated
services using virtualization”
                                                                         as Prof.Dr.




                                                                   105                                http://sites.google.com/site/ijcsis/
                                                                                                      ISSN 1947-5500