| | SLO | ENG | Cookies and privacy

Bigger font | Smaller font

Show document

Title:Metoda planiranja in optimiranja kapacitet dinamičnih strežnih sistemov
Authors:Klampfer, Saša (Author)
Čučej, Žarko (Mentor) More about this mentor... New window
Files:.pdf DR_Klampfer_Sasa_2012.pdf (5,41 MB)
 
Language:Slovenian
Work type:Dissertation (m)
Organization:FERI - Faculty of Electrical Engineering and Computer Science
Abstract:V doktorski disertaciji predstavljamo reševanje problema načrtovanja in planiranja strežnih kapacitet dinamičnih strežnih sistemov s stohastičnimi izbruhi. Skladno z reševanjem omenjenega problema rešujemo tudi problem dinamične rezervacije in redukcije telekomunikacijskih linij. Primarni cilj raziskovalnega dela se navezuje na iskanje optimalnega števila vhodnih telekomunikacijskih linij v Margento strežni sistem mobilnega plačevanja ob minimizaciji števila zavrnjenih transakcijskih zahtev v obdobjih koničnega stohastičnega obremenjevanja. Z iskanjem optimalne rešitve za specifičen scenarij rešujemo še problematiko točne napovedi nabave strojne opreme ter problematiko ekonomsko učinkovitega planiranja. Izhodišče za izpeljavo podmodelov in glavnega modela Margento strežnega sistema nam predstavljajo realni rezultati, ki opisujejo obnašanje v pravem sistemu. Razvita metoda temelji na teoriji funkcij porazdelitev, ki jih kot temeljni matematični element uporabljamo v simulacijskem modelu. V navezi z vključeno optimizacijsko funkcijo, ki deluje v režimu spreminjanja parametra strežnih kapacitet in z ozirom na postavljen prag nam razviti model in optimizacija tvorita predlagano metodo planiranja strežnih kapacitet, katere osnova sta simulacija/emulacija ter matematične funkcije porazdelitev. Z optimizacijo hkrati minimiziramo stroške načrtovanja in dimenzioniranja procesnega centra, strežnega sistema, saj najem vsake vhodne povezave predstavlja dodaten strošek, kot predstavlja dodaten strošek tudi vsaka neizkoriščena telekomunikacijska povezava do strežnega sistema. Razlog za razvoj lastne metode načrtovanja strežnih kapacitet gre iskati v specifičnosti obstoječih rešitev (analitično reševanje, linearne napovedi ipd.), kjer slednje postanejo prekompleksne v primeru stohastičnih obremenitev, velike dinamike pri spremembah obremenitev strežnega sistema, ki jih v regulacijski tehniki pojmujemo kot nihanja v sistemu itd. S predlagano in kasneje predstavljeno metodo smo pokazali še njeno univerzalno uporabnost na področjih procesov in sistemov katerih dogajanje lahko opišemo z uporabo enakih funkcij porazdelitev, kot jih uporabljamo v primeru Margento strežnega sistema. Validacijo metode smo izvedli skladno z realnimi podatki iz realnega sistema, kjer smo število potrebnih telekomunikacijskih povezav dobljenih na osnovi metode primerjali s številom maksimalno hkrati izrabljenih povezav v realnem sistemu (logi). V praksi se lahko pojavita tudi oba robna scenarija, in sicer premalo, oziroma preveč povezav, ki sta nezaželena tako za uporabnike kot tudi za ponudnika določene storitve. S predlagano metodo in lastnim razvitim orodjem (simulatorjem), ki nam predstavlja zgolj pripomoček za analizo različnih scenarijev, želimo dvigniti nivo in kvaliteto storitev ter hkrati iz ekonomskega stališča reducirati in optimirati stroške investicij v nadgradnjo strojne opreme kot tudi najema vhodnih telekomunikacijskih povezav. Optimalno rešitev lahko najdemo z ročnim spreminjanjem parametra strežnih kapacitet (zamudno), oziroma avtomatično, s pomočjo metode avtomatizacije simulacijskih tekov ter avtomatičnega spreminjanja količine strežnih kapacitet. V primeru avtomatičnega iskanja optimuma, razviti simulator, kot orodje in pripomoček, sam generira število simulacijskih tekov in v vsakem izmed njih prilagaja strežno kapaciteto, dokler ne najde ponovljive rešitve v skladu s postavljenim pragom (še sprejemljiv nivo zavrnjenih klicev, transakcij oz. z obzirom na maksimalno še dovoljeno čakalno dobo v čakalni vrsti). V disertaciji podrobno predstavljamo tudi ključne segmente, ki sestavljajo strežni sistem (normalni oz. lažni klic, statistična Gaussova krivulja porazdelitve klicev, mehanizmi sprejemanja in zavračanja klicev, transakcij, upravljanje s kapaciteto vhodnih povezav, naključno proženje klicev itd.) in so hkrati podmodeli predlagane metode. Neposredno primerjavo, s katero pokažemo veljavnost teze na področju planiranja in reševanja že omenjene problema
Keywords:planiranje kapacitet, stohastika, naključnost, simulacija, emulacija, metoda, modeliranje, statistični model, klicalec, transakcija, porazdelitev klicev, strežni sistem, obremenjevanje, normalna porazdelitev, optimizacija, redukcija.
Year of publishing:2012
Publisher:[S. Klampfer]
Source:Maribor
UDC:519.872-048.34:[004.78:621.395](043.3)
COBISS_ID:16001046 Link is opened in a new window
NUK URN:URN:SI:UM:DK:UD4XL1QH
Views:1627
Downloads:133
Metadata:XML RDF-CHPDL DC-XML DC-RDF
Categories:KTFMB - FERI
:
  
Average score:(0 votes)
Your score:Voting is allowed only for logged in users.
Share:AddThis
AddThis uses cookies that require your consent. Edit consent...

Hover the mouse pointer over a document title to show the abstract or click on the title to get all document metadata.

Secondary language

Language:English
Title:Dynamic serving systems capacity planning and optimizing method
Abstract:Within the PhD thesis we represent technique how to precisely solve the capacity planning problem for dynamic serving systems which contains stochastic bursts. Corresponding to this, we at the same time also solve dynamic telecommunication lines reservation or reduction problem. The research work primary aim is focused on searching such an optimal serving system capacity solution which will minimize the rejected transaction attempts number under defined limitation during stochastic peak hour load. By searching the optimal capacity solution we also search the proper and economic solution for hardware equipment purchases. For solving all mentioned problems we propose the new capacity planning and optimizing method based on distribution functions (more precisely – based on normal distribution functions). Issue point for deriving new method was real results recorded within the real Margento Mobile Payment serving system. Regarding such results we pick the proper distribution function, which best describes the peak hour load in the Margento serving system. The whole serving system has been split on many subparts, where each of them has been modeled with proper distribution function. The distribution functions are the fundamental part of derived sub models and models which have been used later in simulation procedure. Corresponding to optimization function which operates in incremental/decremental regime and increments or decrements the serving system capacity parameter regarding to achieve rejected transactions attempts ratio under defined level. Models based on distribution functions and optimization loop creates the new proposed planning method usable also for systems which contain bursts, peak hour load intervals etc. With optimization and new planning method we can dimension the processing server as well as reduce the costs of this, since hiring an input line actually presents quite a substantial cost. The reason why we proposed the new planning method can be found in specifics of existing methods. In very complex systems the analytical methods be-came wasteful and computation time to find proper solution, especially if bursts are presented, dramatically increases. All mentioned existing methods are not proper for our needs, because in all cases the unique linearization level is presented to solve complex problems as stochastic bursts and high dynamic changes in serving process are. Such dynamic changes are well known under regulation scientific area as fluc-tuations within the system. With proposed and later presented method we will show their universality of use on such system and process areas, where the happening can be described with the same distribution functions, which are used in case of Margento serving system. Proposed method validation was performed in accordance with real data obtained from real system. The obtained result of our method was then compared with maximal occupied slots at the specific time (peak hour – logs) in real system. The mutual results comparison will show how precise the proposed method is. In practice we can found also both border scenarios, where serving capacity is too small or in opposite case to big. The first scenario is undesirable for end users; meanwhile the second one is undesirable for service providers which offer specific services. With proposed method and own developed tool (MIMO Simulator) used as experimental tool for testing different scenarios and proving the method correctness we would like to improve the services quality and from economic aspects reduced and optimized investment costs into hardware equipment and lower the costs of hiring input telecommunication connections. The optimal solution can be found manually (slow approach) or automatically by changing (incrementing/decrementing) the serving capacity. In automatic case the developed simulator generates the number of runs and in each of them increments or decrements the capacity parameter, till then, when repeatedly solution is found, regarding to defined limits (rejected transactions ratio must be under
Keywords:capacity planning, stochastic, randomness, simulation, emulation, method, modeling, statistical model, caller, transaction, calls distribution, serving system, load, normal distribution, optimization, reduction


Comments

Leave comment

You have to log in to leave a comment.

Comments (0)
0 - 0 / 0
 
There are no comments!

Back
Logos of partners University of Maribor University of Ljubljana University of Primorska University of Nova Gorica