Another Benchmark for Joining Two Data Frames

In my post yesterday comparing efficiency in joining two data frames, I overlooked the computing cost used to convert data.frames to data.tables / ff data objects. Today, I did the test again with the consideration of library loading and data conversion. After the replication of 10 times in rbenchmark package, the joining method with data.table is almost 10 times faster than the other in terms of user time. Although ff package is claimed to be able to handle large-size data, its efficiency seems questionable.

n <- 1000000
set.seed(2013)
ldf <- data.frame(id1 = sample(n, n), id2 = sample(n / 100, n, replace = TRUE), x1 = rnorm(n), x2 = runif(n))
rdf <- data.frame(id1 = sample(n, n), id2 = sample(n / 100, n, replace = TRUE), y1 = rnorm(n), y2 = runif(n))

library(rbenchmark)
benchmark(replications = 10, order = "user.self",
  # GENERIC MERGE() IN BASE PACKAGE
  merge = merge(ldf, rdf, by = c("id1", "id2")),
  # DATA.TABLE PACKAGE
  datatable = {
    ldt <- data.table::data.table(ldf, key = c("id1", "id2"))
    rdt <- data.table::data.table(rdf, key = c("id1", "id2"))
    merge(ldt, rdt, by = c("id1", "id2"))
  },
  # FF PACKAGE
  ff = {
    lff <- ff::as.ffdf(ldf)
    rff <- ff::as.ffdf(rdf)
    merge(lff, rff, by = c("id1", "id2"))
  },
  # SQLDF PACKAGE
  sqldf = sqldf::sqldf(c("create index ldx on ldf(id1, id2)",
                         "select * from main.ldf inner join rdf on ldf.id1 = rdf.id1 and ldf.id2 = rdf.id2"))
)

#        test replications elapsed relative user.self sys.self user.child
# 2 datatable           10  17.923    1.000    16.605    1.432          0
# 4     sqldf           10 105.002    5.859   102.294    3.345          0
# 1     merge           10 131.279    7.325   119.139   13.049          0
# 3        ff           10 187.150   10.442   154.670   33.758          0