MySQLScalabilitywithUSL
洪斌
LinearScalability?
WhatisScalability?
thecapabilityofasystem,work,orprocesstohandleagrowingamountofwork,oritspotentialtobeenlargedinordertomodatethatgrowth Scalabilityisfunction. ScalabilityLaw •Little’sLaw(1961)•Amdahl’sLaw(1967)•Gustafson’sLaw(1988)•UniversalScalabilityLaw(1993) Queueingtheory •服务请求量量=到达率*驻留留时间(响应时间)•队列列⻓长度=到达率*等待时间•利利⽤用率=到达率*服务时间 Amdahl’sLaw IfanamountofworkNpletedintimeT1onauniprocessor,thesameamountofworkcanpletedintimeTp(2)(1)
(5)(4)
(3) StandardMRM 在有限的p个机器器的⽣生产线，每⼯工作Z段时间就有机器器故障，需要花费S段时间修复，如果多个机器器故障按FIFO顺序修复。
ScalabilityModel •LinearScalability•Costofsharingresources•Diminishingreturnsfromcontention•Negativereturnsfromincoherency ScalabilityZones Contention&Coherency Contention(α) Coherency(β) 含义 共享数据的争⽤用 ⼀一致性的开销 举例例不不同请求更更新相同数据⾏行行内存与缓磁存盘间间的或⼀不一不致同性CPU的 根源 ⽆无法并⾏行行的任务 进程间同步的开销 ⾃自变量量 N-1:假设需要处理理N个进程,最坏场景下有N-1个进程在等待 N*(N-1):假设需要处理理N个进程,每个进程间要与N-1个进 程同步，即N*(N-1) Predict •Predictmaximumscalability•PredictthroughputXmaxatloadNmax DBCapacityPlanning •基准测试估计容量量（时间和成本）•没有完整数据库的负载组成信息•⽆无法准确度量量事务的执⾏行行时间 SteptoApplyUSL
1.选择度量量参数 •Load:QPS/TPS•Concurrency:Thread_running(MySQL)
2.搜集数据 •mysqladmin-i1ext|awk'BEGIN{printf"%5s%5s\n","conn","tput"}/ Threads_running/{run=$4}/Queries/{q=$4-qp;qp=$4;printf"%5d%5d\n",q,run}'
3.整理理数据
4.拟合数据
5.分析结果 Example sample<-read.csv(“8003.tput",sep="")usl<-nls(tput~conn/(1+sigma*(conn-1)+conn*(conn-1)),sample,start=c(sigma=0.1,kappa=0.01))sigma<-coef(usl)['sigma']kappa<-coef(usl)['kappa']u=function(x){y=x/(1+sigma*(x-1)+kappa*x*(x-1))}plot(u,0,max(benchmark$conn)*2,xlab=“Concurrency",col="green",ylab="Throughput",lty=“dashed",add=TRUE)points(benchmark$conn,benchmark$tput) https://kevinbin.shinyapps.io/uslapp/ Conclusions •Scalability是可以被量量化的•线性扩展意味着资源翻倍，负载也翻倍•资源垂直扩展不不意味处理理性能增加，关键是串串⾏行行化⽐比例例。
•即便便极⼩小Coherency也会使Scalability倒退•具备良好Scalability的系统应尽可能避免Contention和 Coherency Reference •HowtoQuantifyScalability(NeilJ.Gunther)•GettingintheZoneforessfulScalability•USLforRpackage•ALittleTriplet•GuerrillaCapacityPlanning•AnalyzingComputerSystemsPerformancewithPerlPDQ “allmodelsarewrong,butsomeareuseful.” –eE.P.Box
WhatisScalability?
thecapabilityofasystem,work,orprocesstohandleagrowingamountofwork,oritspotentialtobeenlargedinordertomodatethatgrowth Scalabilityisfunction. ScalabilityLaw •Little’sLaw(1961)•Amdahl’sLaw(1967)•Gustafson’sLaw(1988)•UniversalScalabilityLaw(1993) Queueingtheory •服务请求量量=到达率*驻留留时间(响应时间)•队列列⻓长度=到达率*等待时间•利利⽤用率=到达率*服务时间 Amdahl’sLaw IfanamountofworkNpletedintimeT1onauniprocessor,thesameamountofworkcanpletedintimeTp
(5)(4)
(3) StandardMRM 在有限的p个机器器的⽣生产线，每⼯工作Z段时间就有机器器故障，需要花费S段时间修复，如果多个机器器故障按FIFO顺序修复。
ScalabilityModel •LinearScalability•Costofsharingresources•Diminishingreturnsfromcontention•Negativereturnsfromincoherency ScalabilityZones Contention&Coherency Contention(α) Coherency(β) 含义 共享数据的争⽤用 ⼀一致性的开销 举例例不不同请求更更新相同数据⾏行行内存与缓磁存盘间间的或⼀不一不致同性CPU的 根源 ⽆无法并⾏行行的任务 进程间同步的开销 ⾃自变量量 N-1:假设需要处理理N个进程,最坏场景下有N-1个进程在等待 N*(N-1):假设需要处理理N个进程,每个进程间要与N-1个进 程同步，即N*(N-1) Predict •Predictmaximumscalability•PredictthroughputXmaxatloadNmax DBCapacityPlanning •基准测试估计容量量（时间和成本）•没有完整数据库的负载组成信息•⽆无法准确度量量事务的执⾏行行时间 SteptoApplyUSL
1.选择度量量参数 •Load:QPS/TPS•Concurrency:Thread_running(MySQL)
2.搜集数据 •mysqladmin-i1ext|awk'BEGIN{printf"%5s%5s\n","conn","tput"}/ Threads_running/{run=$4}/Queries/{q=$4-qp;qp=$4;printf"%5d%5d\n",q,run}'
3.整理理数据
4.拟合数据
5.分析结果 Example sample<-read.csv(“8003.tput",sep="")usl<-nls(tput~conn/(1+sigma*(conn-1)+conn*(conn-1)),sample,start=c(sigma=0.1,kappa=0.01))sigma<-coef(usl)['sigma']kappa<-coef(usl)['kappa']u=function(x){y=x/(1+sigma*(x-1)+kappa*x*(x-1))}plot(u,0,max(benchmark$conn)*2,xlab=“Concurrency",col="green",ylab="Throughput",lty=“dashed",add=TRUE)points(benchmark$conn,benchmark$tput) https://kevinbin.shinyapps.io/uslapp/ Conclusions •Scalability是可以被量量化的•线性扩展意味着资源翻倍，负载也翻倍•资源垂直扩展不不意味处理理性能增加，关键是串串⾏行行化⽐比例例。
•即便便极⼩小Coherency也会使Scalability倒退•具备良好Scalability的系统应尽可能避免Contention和 Coherency Reference •HowtoQuantifyScalability(NeilJ.Gunther)•GettingintheZoneforessfulScalability•USLforRpackage•ALittleTriplet•GuerrillaCapacityPlanning•AnalyzingComputerSystemsPerformancewithPerlPDQ “allmodelsarewrong,butsomeareuseful.” –eE.P.Box
声明：
该资讯来自于互联网网友发布，如有侵犯您的权益请联系我们。