Performance 2002 21 javaservicehanmail net Copyrightc 2002 Java
- Slides: 43
웹 기반 시스템 하에서의 성능(Performance)에 관한 이론적 고찰 2002. 21 자바서비스넷 이원영 javaservice@hanmail. net Copyright(c) 2002 Java. Service. Net 1
1. Performance Concept Copyright(c) 2002 Java. Service. Net 2
What is the performance ? A B http: //www. foo. com/servlet/A http: //www. bar. com/servlet/B 단위응답시간 1 seconds 단위응답시간 5 seconds Copyright(c) 2002 Java. Service. Net 3
Water Pipe Response Time Key Point : 단위시간당 전송된 물의 량 Copyright(c) 2002 Java. Service. Net 5
Performance 측정 Tool - Load. Runner Mercury Interactive Corporation : http: //www. merc-int. com - Microsoft Web Application Test Tool http: //webtool. rte. microsoft. com - Apache Bench $apache/bin/ab - Apache JMeter http: //www. apache. org - IBM AKTools http: //aktools. raleigh. ibm. com Copyright(c) 2002 Java. Service. Net 7
Performance - TPS & Response Mean Time Copyright(c) 2002 Java. Service. Net 9
Performance 의 정의 - Maximum TPS(Transaction Per Second), TPM(… per minute) - 단위시간당 최대 처리가능건수 - Maximum RPS(Request Per Second) - Maximum PPS(Page Per Second) - Active Clients 에 대한 TPS 그래프에서 최대TPS수치 최대 TPS Copyright(c) 2002 Java. Service. Net 10
Tuning 에 관련한 용어 Saturation Point TPS(Transaction/sec) TPS (Transaction Per Second) Saturation Point Light-load zone Heavy-load zone Buckle zone Light-load Heavy load zone Buckle zone A B C Active Clients Copyright(c) 2002 Java. Service. Net 11
Tuning 에 따른 평균응답시간의 변화 Copyright(c) 2002 Java. Service. Net 12
Tuning 에 따른 TPS의 변화 Copyright(c) 2002 Java. Service. Net 13
Tuning 에 따른 TPS/RT의 변화 Copyright(c) 2002 Java. Service. Net 14
Tuning 에 따른 TPS의 변화(2) Throughput(transaction/sec) Saturation Point Tuning Key Effect (83. 33 -65. 22)/65. 22 * 100 28% 의 성능향상 Light-load A Heavy load zone B Buckle zone C Active Clients Copyright(c) 2002 Java. Service. Net 15
Tuning 에 따른 TPS/RT의 변화(3) B A Copyright(c) 2002 Java. Service. Net 16
Performance Tuning Points – cont. . SMP Scaling - Servlet Primitives (1 -CPU, 2 -CPU, 4 -CPU) Copyright(c) 2002 Java. Service. Net 18
Performance Tuning Points – cont. . Vertical scaling with Web. Sphere workload management NOTE: 4 -way 이하에서는 Cloning 의 효과가 거의 없다. App. Server 하나 만으로도 CPU 100%를 사용한다. Copyright(c) 2002 Java. Service. Net 19
Performance Tuning Points – cont. . Machine Cluster Scaling with Network Dispatcher Copyright(c) 2002 Java. Service. Net 20
N/W Line Bandwidth - bps : bit per second (1 byte = 8 bit, 1 Kbyte = 1024 byte) - T 1 : 1. 544 Mbps(1. 5 Mbps) - E 1 : 2. 048 Mbps (2 Mbps) - T 3 : 43. 7361 Mbps (45 Mbps) - OC 3 : 155. 2 Mpbs - 10 Mbps, 100 Mbps - 10 Mbps급: 실전송속도의 60%라고 한다면 실전송속도 6 Mbps = 6*1024(Kbps) = 6*1024/8(KB/sec) = 768(KB/sec) 10 K data전송 : 768(KB/sec) / 10(KB) = 76. 8 TPS 100 Kdata 전송 : 768(KB/sec) / 100(KB) = 7. 68 TPS - 100 Mbps 의 경우, 이것의 10배 향상 - T 1 위 경우, 실전송속도가 80%라고 한다면, 10 K전송시 158. 1(KB/sec) / 10(KB) = 15. 8 TPS 100 K 전송시 : 1. 58 TPS Copyright(c) 2002 Java. Service. Net 21
2. Concurrent Users Copyright(c) 2002 Java. Service. Net 22
K-Internet Banking System 재오픈 2001. 10. 13 (토) Internet F/W A 45 Mbps x 2 L 7(1) Internet Clients 900 -1, 200 sessions per F/W (Max: 7000 per F/W) F/W 900 -1, 200 sessions per F/W (Max: 7000 per F/W) Timeout: 30 min Internet Banking System L 7(2) 1 2 3 4 5 250 user x 3 = 750 users Timeout: 15 min B 1, 200 -1, 400 sessions L 7 600 concurrent sessions per server Timeout : 15 min Inner users C System operators Backend System Which is the concurrent users? - L 4 Switching Device concurrent sessions(2 hours) - F/W sessions(2 hours) - Servlet Http. Session sessions(30 minutes) - Application Login count(10 minutes) - “netstat –n | grep. 80 | grep EST | wc –l” 6 7 IBM RS 6000 SP (4 -way, 2 GB) Copyright(c) 2002 Java. Service. Net - 현재 Request 를 날리고, 응답을 기다리고 있는 사용자 - 현재 해당 시스템을 사용하기 위해 PC앞에 앉아 있는 사용자 (Terminal Users) 23
What’s Concurrent Users ? - L 4 Switching Device concurrent sessions(2 hours) - F/W sessions(2 hours) - Servlet Http. Session sessions(30 minutes) - Application Login count(10 minutes) - “netstat –n | grep. 80 | grep EST | wc –l” Key Point : - 부하량을 측정하기위한 단위 - 즉, 단위시간당 흐른 물의 양 Copyright(c) 2002 Java. Service. Net 24
“Concurrent Users(동시단말사용자)” 어떻게 구할 것인가 ? 특정 시각을 기준으로 10분 전후 모두 한번 이상 호출된, 서로 다른 IP Address 의 개수 Apache access_log 단점 : - Proxy IP Address - e. ND or L 7: 대표IP 대안: è Http. Session ID 10: 30 Concurrent Users : 6명 Copyright(c) 2002 Java. Service. Net 26
Concurrent Users Sample 1 : H-사이트 Copyright(c) 2002 Java. Service. Net 27
Concurrent Users Sample 1 : D-사이트 Copyright(c) 2002 Java. Service. Net 29
How many maximum users ? Example : app 1. jsp 83. 33 1409 120 Maximum TPS : 83. 33 means 833 persons who are hitting every 10 seconds !!! Suppose : Clients request every 10 seconds. Copyright(c) 2002 Java. Service. Net 30
How many maximum users ? Example : app 1. jsp Max Users = Active Clients x 120 120 r = 1, 409 t-r = 8, 591 {t/r} Active Clients Mean Response AC x { t / r } 10 533 188 50 734 681 120 1409 852 160 1871 855 200 2283 876 250 2853 876 t = 10, 000 120 명 Copyright(c) 2002 Java. Service. Net 31
How many maximum users ? Concurrent Users = S. Point x ( t / r ) Concurrent Users 852 120 Appendix F Copyright(c) 2002 Java. Service. Net Active Clients 32
How many maximum users ? C/S Env Example : app 1. jsp Concurrent Users = Active Clients x r = 1, 409 t = 10, 000 { (r+t) / r } Active Clients Mean Response AC x (t/r) AC x { (r+t) / r } 10 533 188 197 50 734 681 731 120 1409 852 972 160 1871 855 1015 200 2283 876 1076 250 2853 876 1126 t + r 120 명 Copyright(c) 2002 Java. Service. Net 33
How many maximum users ? C/S Env. Concurrent Users C/S Concurrent Users = AC x { (r+ t) / r } 972 852 120 Appendix F Active Clients Copyright(c) 2002 Java. Service. Net 34
Acceptable Response Time Copyright(c) 2002 Java. Service. Net 36
3. Performance Measurement In real environment Copyright(c) 2002 Java. Service. Net 38
2 Applications TPS 40 A application B application TPS 40 20 Saturation Point 40 Active Threads [Graph 1] 40 Active Threads [Graph 2] TPS 40 B 10 B 5 10 20 40 Active Threads [Graph 4] TPS 40 A [Graph 3] B 15 10 30 40 Active Threads [Graph 5] TPS 40 B TPS 40 A 30 A 20 20 20 Saturation Point TPS 40 40 A 10 30 40 Active Threads [Graph 6] TPS 40 30 20 20 10 15 10 5 A [Graph 7] B Copyright(c) 2002 Java. Service. Net A B [Graph 8] A B [Graph 9] 39
2 Applications A 40 30 20 10 10 Copyright(c) 2002 Java. Service. Net 20 B 40
2 Applications y T 2 (0, T 2) x + T 1 y T 2 = 1 x Q t 1 y t 2 P t 2 (t 1, t 2) (0, 0) O = (T 1, 0) t 1 T 1 Copyright(c) 2002 Java. Service. Net x 41
3 Applications z T 3 (0, 0, T 3) x t 3 t 1 = y t 2 = z t 3 Q P (t 1, t 2, t 3) (0, 0) (0, T 2, 0) t 2 O T 1 t 1 (T 1, 0, 0) x T 1 + y T 2 + z T 3 T 2 y = 1 x Copyright(c) 2002 Java. Service. Net 42
n- Applications Copyright(c) 2002 Java. Service. Net 43
- Jvm vs clr
- T. trimpe 2003 http://sciencespot.net/
- T.trimpe 2003 http //sciencespot.net/ answers
- Chapter 11 performance appraisal - (pdf)
- Behaviorally anchored rating scale
- Jcids manual
- Java performance counters
- Java rmi performance
- Java io performance
- Java swing form example
- Java import java.util.*
- Import java.awt.*
- Java import java.util.*
- Import java.io.file;
- Import java.util.*
- Import java.util.random
- Java import java.io.*
- Public class
- Java thread import
- Perbedaan antara java swing dan awt adalah
- Import java.awt.event
- Programming language b
- What is rmi and ejb in java
- Odp.net performance monitoring
- Java.net package
- .net vs java
- Import java.net.*;
- Docuware api
- Achmed lach net
- Ado.net vb.net
- Brady and weil 2002
- May 2002 calendar
- 14/4/2002
- 2002-2014
- 08/07/2002
- Rmc 82-2008
- Creating reports in access
- Features of marine insurance
- Identitovigilance loi 4 mars 2002
- Contoh isi borang uma-8
- Samiti sabhya role in odisha
- Flugtag ’88
- Solar eclipse of december 4, 2002
- El decreto 3222 del 2002 se refiere a