Implementation of a new scheduling policy in web servers

الموضوعات

علوم تطبيقية وتكنولوجية

الملخص

A widely encountered problem in web servers over the Internet is the long responsetime. It is possible to reduce the response time of requests at a web server by simplychanging the order in which we schedule the requests. Recently, the Shortest-Remaining-Processing-Time (SRPT) has been proposed for scheduling requests in web servers. TheSRPT assumes that the response time of the requested Öle is strongly proportional to itssize. However, depending only on the size of the Öle for determining the priority of therequest is not enough, since it doesnít take into consideration the client-server interactionthrough the Internet, where web servers are mainly used. In the Internet, the clients aregeographically dispersed which presents high diversity in path bandwidth, round-trip timeand packet loss characteristics. To account for these parameters, this thesis proposes a newscheduling policy for processing static HTTP requests in web servers that better estimatesthe response time. We call this policy, Shortest-Remaining-Response-Time (SRRT).Our approach beneÖts from the TCP implementation to capture useful scheduling infor-mation about the interaction between the server and the client through the network. TheSRRT prioritizes requests based on a combination of the current Round-Trip-Time (RTT),TCP window size and the size of what remains of the requested Öle. The requests whichhave the shortest estimated remaining response time receive higher priorities.The implementation is done at the kernel level for controlling the order in which socketbu§ers are drained into the network. Our experiment uses the Linux operating system andthe Apache web server. In the experiment the requests are generated by the Scalable URLRequest Generator (SURGE) workload generator, and the WAN is represented by NetworkEmulation (netem).We compare SRRT to SRPT and processor-sharing (PS) policies. SRRT and SRPTshow an improvement over PS. However, the SRRT shows the best improvement in thexi mean response time. SRRT gives an average improvement of about 7.5% over SRPT forboth 10Mbps and 100Mbps links and under all loads. For 10Mbps link, the maximumimprovement of SRRT over SRPT is 13.2%. While for the 100Mbps link the maximumimprovement is 11.6%.This improvement comes at a negligible expense in response time for long requests. Wefound that under 100Mbps link, only 1.5% of long requests have longer response times. Thelongest request under SRRT has an increase in response time by a factor 1.7 over PS. For10Mbps link, only 2.4% of requests are penalized, and SRRT increases the longest requesttime by a factor 2.2 over PS

روابط وملفات

التعريف والنوع

رقم الوثيقة

fa484bf4-1019-48f0-af6d-77f2d78ad489

رقم العقد

0

نوع الوسائط

Crawler

نوع المحتوى

الرسائل العلمية

صيغة المصدر

رسائل ماجيستير

نوع الملف

pdf text

أسماء الملفات

57797_1.pdf

بيانات النشر

ألقاب المؤلفين

[{"name_ar":"Ahmad S. Al-Saídeh","title_ar":"اعداد","title_en":"Preparation"},{"name_ar":"عدنان يحيي","title_ar":"اشراف","title_en":"Supervision"}]

اللغة

English

المصدر والدورية

اسم المصدر

Implementation of a new scheduling policy in web servers

المحتوى والصفحات

عدد الصفحات

0

إشراف وإعداد

الإشراف

عدنان يحيي

الإعداد

Ahmad S. Al-Saídeh

الاقتباسات الببليوغرافية