Yaksha A SelfTuning Controller for Managing the Performance
- Slides: 29
Yaksha: A Self-Tuning Controller for Managing the Performance of 3 -Tiered Web Sites Abhinav Kamra, Vishal Misra CS Department Columbia University Erich Nahum IBM TJ Watson Research Center
Dynamic Content Online shopping n Amazon, Best. Buy News snippets n http: //news. google. com/ Current weather conditions Real-time stock tickers
Dynamic Content Generation http Web Server App Server Database Server 3 -Tier Structure: n n n Web Server: Static web pages App Server: CGI / Java servlets Database Server: Backend Data Store
Major Problems Overloaded Web Sites: n n The “Slashdot Effect” Unanticipated load causes site to crash Unresponsive Web Sites: n n The “Abandoned Shopping Cart’’ Unacceptable delays lead to reduced usage
Throughput Admission Control Ideal Actual Load To prevent overload, perform admission control: n n n Notion of capacity in the system Identify the job ahead of time & amount of work generated Only let jobs in if they won’t overload system Once you reach full capacity: n n Make jobs wait Drop jobs
Why Self-Tuning ? Parameter Setting n n Lots of experimentation Workload characterization Re-done for every system change
Outline Motivation & Background The ‘Yaksha’ Controller n n Architecture Modeling Design Self-Tuning Experimental Environment Experimental Results Summary and Conclusions
The ‘Yaksha’ Controller Architecture Clients http Yaksha Web Server App Server Database Server Intercepts HTTP requests n Decides whether to accept or reject new connections Maintains several measurement-based estimates: n n n Per connection Response and Sojourn times Per customer-class based estimates Per query-type based estimates
Modeling + S Controller Web Server – Reference Input n n = Desired Response/Sojourn times = Incoming job acceptance probability
Modeling System Abstraction n n M/GI/1 Processor Sharing Queue Linearization approximation Open loop transfer function n
Design Proportional Integral (PI) Control n n Zero steady state error Closed loop transfer function n
Design (contd. ) Setting system parameters n Fix controller time constant to 10 sec n Fix phase margin at 45 degrees Bilinear transform to convert to digital form
Self-Tuning ‘Pure gain’ open loop transfer function Effective arrival rate ‘Tuned’ transfer function Running average for pa
Parameter Setting Parameters w/o Self-Tuning n n n Expected input rate Expected connection drop rate Target response time Parameters with Self-Tuning n Target response time
Outline Motivation & Background The ‘Yaksha’ Controller Experimental Environment n n Setup & Methodology Software & Hardware Experimental Results Summary and Conclusions
Experimental Setup SQL Workload Generator Web/App Server Controller Lightweight Proxy Database Server
Emulated Clients http Tomcat SQL My. SQL Remote Browser Emulator n Session duration n Think time n Markov model Load is a function of the number of clients
Software Workload Generator Web/App Server Controller Lightweight Proxy Workload Generator TPC-W 1. 0. 1 Lightweight Proxy Tinyproxy 1. 6. 1 Web/App Server Tomcat 4. 1. 27 Database Server My. SQL 4. 1. 0 SQL Database Server
Hardware TPC-W Client http Tinyproxy/ Tomcat CPU Intel Pentium 1. 7 GHz Memory 512 MB Disk 12 GB, 12 ms, 5400 RPM Network 100 Mbps Ethernet SQL My. SQL
Outline Motivation & Background The ‘Yaksha’ Controller Experimental Environment Experimental Results n n Response time control Throughput control Self-tuning Model validation Summary and Conclusions
Results: Response time control
Results: Throughput control
Results: Self-tuning
Results: Self-tuning
Results: Model Validation
Summary & Future Work Presented the ‘Yaksha’ Control System n n PI admission control for http connections Overload prevention Response time bounds Self-Tuning Control Future Work n Throughput maximization
Thank You!
Related Work Admission Control for Static Content Web Servers: n Bhatti 99, Li 00, Voigt 01, Pradhan 02 n Provide throughput/response time/BW guarantees Control n Tarek 01, Tarek 02, Hellerstein 01, Hellerstein 02, Welsh 03 n Control theory for resource management n Admission control for Apache, Lotus notes Dynamic Content: n Dynaserver project at Rice n TPC-W Benchmarks
Results: Throughput control - Pa
- Managing employee performance and reward
- Jag har nigit för nymånens skära
- Inköpsprocessen steg för steg
- Skivepiteldysplasi
- Strategi för svensk viltförvaltning
- Formula varians
- Tack för att ni har lyssnat
- Datorkunskap för nybörjare
- Läkarutlåtande för livränta
- Klassificeringsstruktur för kommunala verksamheter
- Ekologiskt fotavtryck
- Påbyggnader för flakfordon
- Tack för att ni lyssnade
- Tobinskatten för och nackdelar
- Egg för emanuel
- Atmosfr
- Verifikationsplan
- Vilotidsbok
- Rutin för avvikelsehantering
- Vanlig celldelning
- Presentera för publik crossboss
- Fuktmätningar i betong enlig rbk
- Kontinuitetshantering i praktiken
- Myndigheten för delaktighet
- Var 1721 för stormaktssverige
- Tack för att ni har lyssnat
- Boverket ka
- Referatmarkeringar
- Varför kallas perioden 1918-1939 för mellankrigstiden
- Karttecken brunn