share: Create a shared object

In Jiefei-Wang/SharedObject: Sharing R objects across multiple R processes without memory duplication

Create a shared object

Description

This function will create a shared object for the object x. The behavior of the shared object is exactly the same as x, but the data of the shared object is allocated in the shared memory space. Therefore, a shared object can be easily exported to the other R workers without duplicating the data, which can reduce the memory consumption and the overhead of data transmission.

Usage

share(x, ...)

## S4 method for signature 'ANY'
share(
  x,
  ...,
  copyOnWrite,
  sharedSubset,
  sharedCopy,
  sharedAttributes,
  mustWork,
  minLength
)

Arguments

x	An R object that will be shared, see details.
...	For generalization purpose.
copyOnWrite, sharedSubset, sharedCopy	The parameters controlling the behavior of a shared object, see details.
sharedAttributes	Whether to share the attributes of the object x (default TRUE). Note that attribute class and names will never be shared.
mustWork	Whether to throw an error if x is not sharable (e.g. x is a function). This parameter has no effect on the object's attributes and S4 object.
minLength	The minimum length of a shared object(default 3). If length(x) is smaller than the minimum length, it would not be shared. This parameter can be used to reduce the memory fragmentation.

Details

The function returns a shared object corresponding to the argument x if it is sharable. There should be no different between x and the return value except that the latter one is shared. The attributes of x will also be shared if possible.

Supported types

For the basic R type, the function supports raw, logical,integer, double, complex. character can be shared, but sharing a character is beneficial only when there are a lot repetitions in the elements of the vector. Due to the complicated structure of the character vector, you are not allowed to set the value of a shared character vector to a value which haven't presented in the vector before. It is recommended to treat a shared character vector as read-only.

For the container, the function supports list, pairlist and environment. Note that sharing a container is equivalent to share all elements in the container, the container itself will not be shared.

The function share is an S4 generic. The default share method works for most S3/S4 objects. Therefore, there is no need to define a S4 share method for each S3/S4 class unless the S3/S4 class has a special implementation (e.g. on-disk data). The default method will share all slots the object contains and the object itself if possible. No error will be given if any of these objects are not sharable and they will be kept unchanged.

Behavior control

The behavior of a shared object can be controlled through three parameters: copyOnWrite, sharedSubset and sharedCopy.

copyOnWrite determines Whether a shared object needs to be duplicated when the data of the shared object is changed. The default value is TRUE, but can be altered by passing copyOnWrite = FALSE to the function. This parameter can be used to let workers directly write the result back to a shared object.

Please note that the no-copy-on-write feature is not fully supported by R. When copyOnWrite is FALSE, a shared object might not behaves as one expects. Please refer to the example code to see the exceptions.

sharedSubset determines whether the subset of a shared object is still a shared object. The default value is FALSE, and can be changed by passing sharedSubset = TRUE to the function

At the time of writing, The shared subset feature will cause an unnecessary memory duplication in R studio. Therefore, for the performance consideration, it is recommended to keep the feature off in R studio.

sharedCopy determines whether the object is still a shared object after the duplication. Note that it must be used with copyOnWrite = TRUE. Otherwise, the shared object will never be duplicated. The default value is FALSE.

Value

A shared object

Examples

## For vector
x <- runif(10)
so <- share(x)
x
so

## For matrix
x <- matrix(runif(10), 2, 5)
so <- share(x)
x
so

## For data frame
x <- as.data.frame(matrix(runif(10), 2, 5))
so <- share(x)
x
so

## export the object
library(parallel)
cl <- makeCluster(1)
clusterExport(cl, "so")
## check the exported object in the other process
clusterEvalQ(cl, so)

## close the connection
stopCluster(cl)

## Copy on write
x <- runif(10)
so1 <- share(x, copyOnWrite = TRUE)
so2 <- so1
so2[1] <- 10
## so1 is unchanged since copy-on-write feature is on.
so1
so2

## No copy on write
so1 <- share(x, copyOnWrite = FALSE)
so2 <- so1
so2[1] <- 10
#so1 is changed
so1
so2

## Flaw of no-copy-on-write
## The following code changes the value of so1,
## highly unexpected! Please use with caution!
-so1
so1
## The reason is that the minus function tries to
## duplicate so1 object, but the duplication function
## will return so1 itself, so the values in so1 get changed.

Jiefei-Wang/SharedObject documentation built on June 8, 2024, 1:35 a.m.