Raamistik pilvel põhinevate veebirakenduste skaleeruvuse ja jõudluse kontrollimiseks
Date
2012
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Tartu Ülikool
Abstract
Antud magistritöö uurib võimalusi, kuidas kasutada veebirakendust MediaWiki, mida kasutatakse Wikipedia rakendamiseks, ja kuidas kasutada antud teenust mitme serveri peal nii, et see oleks kõige optimaalsem ja samas kõik veebikülastajad saaks teenusele ligi mõistliku ajaga. Amazon küsib raha pilves toimivate masinate ajalise kasutamise eest, ümardades pooleldi kasutatud tunnid täistundideks. Antud töö sisaldab vahendeid kuidas mõõta pilves olevate serverite jõudlust ning võimekust ja skaleerida antud veebirakendust.
Amazon EC2 pilvesüsteemis on võimalik kasutajatel koostada virtuaalseid tõmmiseid operatsiooni süsteemidest, mida saab pilves rakendada XEN virtualiseerimise keskkonnas, kui eraldiseisvat serverit. Antud virtuaalse tõmmise peale sai paigaldatud tööks vaja minev keskkond, et koguda andmeid serverite kasutuse kohta ja võimaldada platvormi, mis lubab dünaamiliselt ajas lisada servereid ja eemaldada neid.
Magistritöö uurib Amazon EC2 pilvesüsteemi kasutusvõimalusi, mille hulka kuulub Auto Scale, mis aitab skaleerida pilves kasutatavaid rakendusi horisontaalselt. Amazon pilve kasutatakse antud töös MediaWiki seadistamiseks ja suuremahuliste eksperimentide rakendamiseks. Vajalik on teha palju optimiseerimisi ja seadistamisi, et suurendada teenuse läbilaske võimsust. Antud töö raames loodud raamistik aitab mõõta serverite kasutust, kogudes andmeid protsessori, mälu ja võrgu kasutamise kohta. See aitab leida süsteemis olevaid kitsaskohti, mis võivad põhjustada süsteemi olulist aeglustumist.
Antud töö raames sai tehtud erinevaid teste, et selgitada välja parim võimalik paigutus ja seadistus. Saavutatud seadistust kontrolliti hiljem 2 suuremahulise eksperimentiga, mis kestis üks päev ja mille käigus tekitati 22 miljonit päringut, leidmaks kuidas raamistik võimaldab teenust pilves skaleerida ülesse päringute arvu tõusmisel ja vähendada servereid, kui päringute arv väheneb. Ühes eksperimendis kasutati optimaalset heuristikat, et selgitada välja optimaalne serverite arv, mida on vaja pilves rakendada. Teine eksperimentidest kasutas Amazon Auto Scale teenust, mis kasutas serverite keskmist protsessori kasutamist, et selgitada välja, kas pilves on vaja servereid lisada või eemaldada. Antud eksperimendid näitavad selgelt, et kasutades dünaamilist arvu servereid, olenevalt päringute arvust, on võimalik teenuse üleval hoidmiseks säästa raha.
Network usage and bandwidth speeds have increased massively and vast majority of people are using Internet on daily bases. This has increased CPU utilization on servers meaning that sites with large visits are using hundreds of computers to accommodate increasing traffic rates to the services. Making plans for hardware ordering to upgrade old servers or to add new servers is not a straightforward process and has to be carefully considered. There is a need to predict traffic rate for future usage. Buying too many servers can mean revenue loss and buying too few servers can result in losing clients. To overcome this problem, it is wise to consider moving services into virtual cloud and make server provisioning as an automatic step. One of the popular cloud service providers, Amazon is giving possibility to use large amounts of computing power for running servers in virtual environment with single click. They are providing services to provision as many servers as needed to run, depending how loaded the servers are and whatever we need to do, to add new servers or to remove existing ones. This will eliminate problems associated with ordering new hardware. Adding new servers is an automatic process and will follow the demand, like adding more servers for peak hours and removing unnecessary servers at night or when the traffic is low. Customer pays only for the used resources on the cloud. This thesis focuses on setting up a testbed for the cloud that will run web application, which will be scaled horizontally (by replicating already running servers) and will use the benchmark tool for stressing out the web application, by simulating huge number of concurrent requests and proper load-balancing mechanisms. This study gives us a proper picture how servers in the cloud are scaled and whole process remains transparent for the end user, as it sees the web application as one server. In conclusion, the framework is helpful in analyzing the performance of cloud based applications, in several of our research activities.
Network usage and bandwidth speeds have increased massively and vast majority of people are using Internet on daily bases. This has increased CPU utilization on servers meaning that sites with large visits are using hundreds of computers to accommodate increasing traffic rates to the services. Making plans for hardware ordering to upgrade old servers or to add new servers is not a straightforward process and has to be carefully considered. There is a need to predict traffic rate for future usage. Buying too many servers can mean revenue loss and buying too few servers can result in losing clients. To overcome this problem, it is wise to consider moving services into virtual cloud and make server provisioning as an automatic step. One of the popular cloud service providers, Amazon is giving possibility to use large amounts of computing power for running servers in virtual environment with single click. They are providing services to provision as many servers as needed to run, depending how loaded the servers are and whatever we need to do, to add new servers or to remove existing ones. This will eliminate problems associated with ordering new hardware. Adding new servers is an automatic process and will follow the demand, like adding more servers for peak hours and removing unnecessary servers at night or when the traffic is low. Customer pays only for the used resources on the cloud. This thesis focuses on setting up a testbed for the cloud that will run web application, which will be scaled horizontally (by replicating already running servers) and will use the benchmark tool for stressing out the web application, by simulating huge number of concurrent requests and proper load-balancing mechanisms. This study gives us a proper picture how servers in the cloud are scaled and whole process remains transparent for the end user, as it sees the web application as one server. In conclusion, the framework is helpful in analyzing the performance of cloud based applications, in several of our research activities.