options(width = 80)
library(data.table)
library(ggplot2)

dt <- data.table(read.table("webcachesim.log"))
setnames(dt,colnames(dt),c("t","type","size","param","h","r","b"))

dt <- dt[,list(h=max(h)),by=list(type,size,param,r)]
#    dt[,list(h1=max(h),h2=min(h)),by=list(type,size,param,r)][h1!=h2]

setkey(dt,type,size,param)

dt[,ohr:=h/r]

# opt param
dt2 <- dt[type=="ExpLRU"]
dt3 <- dt2[,list(param,ohr,mohr=max(ohr)),by=size]
dt4 <- dt3[ohr==mohr]

pl <- ggplot(dt4,aes(size,param))+
    geom_line()+
        scale_x_continuous("Cache size",breaks=c(30+log2(50),30+log2(100),30+log2(200),30+log2(400)),labels=c("50GB","100GB","200GB","400GB"))+
            scale_y_continuous("Optimal c param",breaks=(13:22),labels=paste(2^((13:22)-10),"KB"))+
    theme_grey(base_size = 22)

png(paste("/tmp/explru-param.png",sep=""),500,300)
print(pl)
dev.off()

# (log-log) linear fit
fit <- lm(param ~ size, data = dt4)

bbase <- fit$coefficients[1]*-1
rrate <- fit$coefficients[2]

dt4
linfn <- function(s) s^rrate/2^bbase
log2(linfn(2^(dt4$size)))


# filter param scaling

dt2 <- dt[type=="Filter"]

dt3 <- dt2[,list(param,ohr,mohr=max(ohr)),by=size]

pl <- ggplot(dt3[ohr==mohr],aes(size,param))+
    geom_line()+
        scale_x_continuous("Cache size",breaks=c(30+log2(50),30+log2(100),30+log2(200),30+log2(400)),labels=c("50GB","100GB","200GB","400GB"))+
    scale_y_continuous("Optimal N param",breaks=(seq(4,50,by=4)),labels=seq(4,50,by=4))+
    theme_grey(base_size = 22)

png(paste("/tmp/filter-param.png",sep=""),500,300)
print(pl)
dev.off()


# hit ratio graphs

dt5 <- dt[,list(ohr=max(ohr)),by=list(size,type)]

lbreaks=c("ExpLRU","Filter","LRU")
llabels=c("Exp-size","Filter","LRU")

dt6 <- dt5[type %in% lbreaks]
dt6[,ttype:=factor(type,levels=lbreaks,labels=llabels,ordered=TRUE)]
dt6[,type2:=factor(type,levels=lbreaks,ordered=TRUE)]

pl <- ggplot(dt6,aes(size,ohr,color=ttype))+
    geom_line(size=1.5)+
    scale_x_continuous("Cache size",breaks=c(30+log2(50),30+log2(100),30+log2(200),30+log2(400)),labels=c("50GB","100GB","200GB","400GB"))+
    scale_y_continuous("Hit ratio")+
    scale_colour_brewer("",palette="Set1")+
    theme_grey(base_size = 22)


png(paste("/tmp/hitratio-cap.png",sep=""),500,300)
print(pl)
dev.off()


# sensititivy graphs

dt2 <- dt[type=="Filter"]
setkey(dt2,ohr)
setkey(dt2,param)

pl <- ggplot(dt2,aes(param,ohr,color=factor(size)))+
    geom_line(size=1.5)+
        scale_x_continuous("Admit after N requests",breaks=(1:10)*10)+
    scale_y_continuous("Hit Ratio",breaks=c(0.75,0.8,0.85,.9,.95))+
        scale_colour_brewer("",palette="Set1",breaks=rev(dt2[size!=38.325,unique(size)]),
                        labels=rev(paste(
                            dt2[size!=38.325,round(2^(unique(size)-30))],
                                "GB")))+
    theme_grey(base_size = 22)+
    coord_cartesian(ylim=c(0.75,0.95))

png(paste("/tmp/Filter-sensitivity.png",sep=""),500,400)
print(pl)
dev.off()


dt2 <- dt[type=="ExpLRU"]
setkey(dt2,ohr)
setkey(dt2,param)
pl <- ggplot(dt2,aes(param,ohr,color=factor(size)))+
    geom_line(size=1.5)+
        scale_x_continuous("Exp param c",breaks=c(10,12,14,16,18,20,22),labels=c("1KB","4KB","16KB","64KB","256KB","1MB","4MB"))+
    scale_y_continuous("Hit Ratio",breaks=(1:50)/50)+
    scale_colour_brewer("",palette="Set1",breaks=rev(dt2[size!=38.325,unique(size)]),
                        labels=rev(paste(
                            dt2[size!=38.325,round(2^(unique(size)-30))],
                                "GB")))+
    theme_grey(base_size = 22) +
        coord_cartesian(ylim=c(0.85,0.98))
png(paste("/tmp/ExpLRU-sensitivity.png",sep=""),500,400)
print(pl)
dev.off()