# Run SMA(20) batch over a panel of assets, serial vs parallel::mclapply, with # the speedup. mclapply forks on Unix; on Windows it runs serially. library(wickra) source("_common.R") args <- commandArgs(trailingOnly = TRUE) assets <- if (length(args) >= 1) as.integer(args[1]) else 200L bars <- if (length(args) >= 2) as.integer(args[2]) else 5000L panel <- lapply(seq_len(assets), function(a) synthetic_prices(bars, start = 50 + a * 0.1)) run_one <- function(p) { s <- Sma(20); r <- batch(s, p); r[length(r)] } t0 <- Sys.time(); invisible(vapply(panel, run_one, numeric(1))) serial <- as.numeric(Sys.time() - t0, units = "secs") # mclapply forks on Unix; Windows has no fork, so it must run on a single core. cores <- if (.Platform$OS.type == "windows") 1L else max(1L, parallel::detectCores()) t0 <- Sys.time(); invisible(parallel::mclapply(panel, run_one, mc.cores = cores)) par <- as.numeric(Sys.time() - t0, units = "secs") cat(sprintf("%d assets x %d bars, SMA(20) batch:\n", assets, bars)) cat(sprintf(" serial %8.1f ms\n", serial * 1000)) cat(sprintf(" parallel %8.1f ms (%.1fx speedup, %d cores)\n", par * 1000, serial / max(par, 1e-9), cores))