W3Cschool
恭喜您成為首批注冊用戶
獲得88經(jīng)驗值獎勵
You want to send and receive large arrays of contiguous data across a network connec‐tion, making as few copies of the data as possible.
The following functions utilize memoryviews to send and receive large arrays:
def send_from(arr, dest):
view = memoryview(arr).cast(‘B')while len(view):
nsent = dest.send(view)view = view[nsent:]
def recv_into(arr, source):
view = memoryview(arr).cast(‘B')while len(view):
nrecv = source.recv_into(view)view = view[nrecv:]
To test the program, first create a server and client program connected over a socket.In the server:
>>> from socket import *
>>> s = socket(AF_INET, SOCK_STREAM)
>>> s.bind(('', 25000))
>>> s.listen(1)
>>> c,a = s.accept()
>>>
In the client (in a separate interpreter):
>>> from socket import *
>>> c = socket(AF_INET, SOCK_STREAM)
>>> c.connect(('localhost', 25000))
>>>
Now, the whole idea of this recipe is that you can blast a huge array through the con‐nection. In this case, arrays might be created by the array module or perhaps numpy.For example:# Server>>> import numpy>>> a = numpy.arange(0.0, 50000000.0)>>> send_from(a, c)>>>
In data-intensive distributed computing and parallel programming applications, it’s notuncommon to write programs that need to send/receive large chunks of data. However,to do this, you somehow need to reduce the data down to raw bytes for use with low-level network functions. You may also need to slice the data into chunks, since mostnetwork-related functions aren’t able to send or receive huge blocks of data entirely allat once.One approach is to serialize the data in some way—possibly by converting into a bytestring. However, this usually ends up making a copy of the data. Even if you do thispiecemeal, your code still ends up making a lot of little copies.
This recipe gets around this by playing a sneaky trick with memoryviews. Essentially, amemoryview is an overlay of an existing array. Not only that, memoryviews can be castto different types to allow interpretation of the data in a different manner. This is thepurpose of the following statement:view = memoryview(arr).cast(‘B')
It takes an array arr and casts into a memoryview of unsigned bytes.In this form, the view can be passed to socket-related functions, such as sock.send()or send.recv_into(). Under the covers, those methods are able to work directly withthe memory region. For example, sock.send() sends data directly from memorywithout a copy. send.recv_into() uses the memoryview as the input buffer for thereceive operation.The remaining complication is the fact that the socket functions may only work withpartial data. In general, it will take many different send() and recv_into() calls totransmit the entire array. Not to worry. After each operation, the view is sliced by thenumber of sent or received bytes to produce a new view. The new view is also a memoryoverlay. Thus, no copies are made.One issue here is that the receiver has to know in advance how much data will be sentso that it can either preallocate an array or verify that it can receive the data into anexisting array. If this is a problem, the sender could always arrange to send the size first,followed by the array data.
Copyright©2021 w3cschool編程獅|閩ICP備15016281號-3|閩公網(wǎng)安備35020302033924號
違法和不良信息舉報電話:173-0602-2364|舉報郵箱:jubao@eeedong.com
掃描二維碼
下載編程獅App
編程獅公眾號
聯(lián)系方式:
更多建議: