Speed up the write operation by bulk calling read.

Bulk check the ACKs from the server every 32MB
(or every write request). This way you gain speed
but also making sure not to get the error too late
in a large transfer.
This works for smaller files too, since there is a
cleanup routine being called when the file has been transfered.

paramiko/sftp_file.py

@@ -21,6 +21,7 @@
 """
 
 from binascii import hexlify
+from collections import deque
 import socket
 import threading
 import time
@@ -54,6 +55,7 @@ def __init__(self, sftp, handle, mode='r', bufsize=-1):
         self._prefetch_data = {}
         self._prefetch_reads = []
         self._saved_exception = None
+        self._reqs = deque()
 
     def __del__(self):
         self._close(async=True)
@@ -163,12 +165,14 @@ def _read(self, size):
     def _write(self, data):
         # may write less than requested if it would exceed max packet size
         chunk = min(len(data), self.MAX_REQUEST_SIZE)
-        req = self.sftp._async_request(type(None), CMD_WRITE, self.handle, long(self._realpos), str(data[:chunk]))
-        if not self.pipelined or self.sftp.sock.recv_ready():
-            t, msg = self.sftp._read_response(req)
-            if t != CMD_STATUS:
-                raise SFTPError('Expected status')
-            # convert_status already called
+        self._reqs.append(self.sftp._async_request(type(None), CMD_WRITE, self.handle, long(self._realpos), str(data[:chunk])))
+        if not self.pipelined or (len(self._reqs) > 100 and self.sftp.sock.recv_ready()):
+            while len(self._reqs):
+                req = self._reqs.popleft()
+                t, msg = self.sftp._read_response(req)
+                if t != CMD_STATUS:
+                    raise SFTPError('Expected status')
+                # convert_status already called
         return chunk
 
     def settimeout(self, timeout):