Effective servers

The main goal of this task is to implement simple server applications in two languages (C/C++ and Java) and optimize its performance in order to be able to handle many clients and their requests.

Minor goals: get familiar with documentation and source code of used platforms, application of the knowledge gained on lectures.

Write an TCP + protobuf based server application, which will count unique words in data sent by clients. The server must meet the following requirements:

• All TCP transfers are composed of two parts: message length (4 bytes, big endian, unsigned int) + serialized protobuf message.
• Clients send data using Request_PostWords protobuf message. Data are in plain text format encoded in UTF-8 and compressed by gzip method.
• The server counts unique words. On startup, the counter equals zero.
• The server keeps records of words sent by clients. The unique word counter is increased by one for each new word, which is not in records yet.
• If a Request_GetCount arrives, the server will answer actual value of unique count and reset it.
• The server must be able to handle a large number of simultaneously connected clients (approximately 100).
• The server has to respond to every client Request message with Response message. Status field in response message is required.

You can use any hardware platform and operating system.

Use the following schema for communication with our clients:

syntax = "proto3";

option java_multiple_files = true;
option java_package = "cz.cvut.fel.esw.server.proto";

message Request {
  message GetCount {};
  message PostWords {
    bytes data = 1;
  }
  oneof msg {
    GetCount getCount = 1;
    PostWords postWords = 2;
  }
}

message Response {
  enum Status { OK = 0; ERROR = 1;};

  Status status = 1; // Always present
  int32 counter = 2;
  string errMsg = 3;
}

You can use any library written in C/C++ (no other languages please).

You can use any library written in Java (you are not allowed to use libraries which are using JNI except libraries which are using functionality implemented in JDK).

The performance of server will be measured by a test program, which will send a bigger amount of data, and then ask for the current value of unique words. Total time from the first Request message send to the counter value reception will be measured.

Solutions twice faster than example server below will be accepted. The reference solution is measured on a computer in the laboratory.

Easy server implementation in Python language, which meets the requirements (except performance), is shown below. You can use this implementation as a reference for your solution. Let's suppose that you saved given proto schema as esw_server.proto and generated python files by using protoc –python_out=. esw_server.proto, then you can use the following code.

#!/usr/bin/env python3

import asyncio
import gzip
import struct

import esw_server_pb2

words = {}


def send_message(writer, message):
    data = message.SerializeToString()
    data_size = struct.pack("!I", len(data))
    writer.write(data_size + data)


async def receive_message(reader, message_type):
    try:
        buffer = await reader.readexactly(4)
    except asyncio.IncompleteReadError as irerr:
        if len(irerr.partial) > 0:
            raise ValueError("Message length transmission incomplete")
        return None
    data_size, = struct.unpack("!I", buffer)
    try:
        data = await reader.readexactly(data_size)
    except asyncio.IncompleteReadError:
        raise ValueError("Data transmission incomplete")

    message = message_type()
    message.ParseFromString(data)
    return message


async def handle_connection(reader, writer):
    global words
    global loop

    # Client may ask multiple requests
    while True:
        request = await receive_message(reader, esw_server_pb2.Request)
        if request is None:
            break
        response = esw_server_pb2.Response()

        if request.HasField("getCount"):
            response.status = esw_server_pb2.Response.OK
            response.counter = len(words)
            send_message(writer, response)
            words = {}
        elif request.HasField("postWords"):
            text = gzip.decompress(request.postWords.data).decode("utf-8")
            for word in text.split():
                words[word] = 1;
            response.status = esw_server_pb2.Response.OK
            send_message(writer, response)
        else:
            raise Exception("Request broken")


loop = asyncio.get_event_loop()
coro = asyncio.start_server(handle_connection, '', 8080, loop=loop)
server = loop.run_until_complete(coro)

print('Serving on {}'.format(server.sockets[0].getsockname()))
loop.run_forever()

You can use the following code to test your server locally. Let say that we saved the following code into tester.py, then we can run python3 tester.py to test trivial functionality of your solution.

#!/usr/bin/env python3

import gzip
import struct
import socket
import sys

import esw_server_pb2


class Client:
    def __init__(self, host, port):
        self.connection = socket.socket(socket.AF_INET6, socket.SOCK_STREAM)
        self.connection.connect((host, port))

    def send_message(self, pb_message):
        data = pb_message.SerializeToString()
        data_size = struct.pack("!I", len(data))
        self.connection.sendall(data_size + data)

    def receive_message(self, messageType):
        data_size, = struct.unpack("!I", self.connection.recv(4))
        data = b''
        while data_size:
            buffer = self.connection.recv(data_size)
            if not buffer:
                raise ValueError("Data transmission incomplete")
            data += buffer
            data_size -= len(buffer)
        message = messageType()
        message.ParseFromString(data)
        return message

    def get_count(self):
        request = esw_server_pb2.Request()
        request.getCount.CopyFrom(esw_server_pb2.Request.GetCount())
        self.send_message(request)
        response = self.receive_message(esw_server_pb2.Response)
        if response.status == esw_server_pb2.Response.OK:
            return response.counter
        else:
            raise ValueError("Status not OK: " + response.errMsg)

    def post_words(self, text):
        request = esw_server_pb2.Request()
        request.postWords.CopyFrom(esw_server_pb2.Request.PostWords())
        request.postWords.data = gzip.compress(text.encode("utf-8"))
        self.send_message(request)
        response = self.receive_message(esw_server_pb2.Response)
        if response.status != esw_server_pb2.Response.OK:
            raise ValueError("Status not OK: " + response.errMsg)


host = sys.argv[1] if len(sys.argv) > 1 else "localhost"
port = int(sys.argv[2]) if len(sys.argv) > 2 else 8080
client = Client(host, port)

print(client.get_count())
client.post_words("two words")
client.post_words("three words words")
print(client.get_count())

For a testing use web application, where you put IP address of your server (any public IP, local IPs in KN:E-311 and KN:E-s109 are allowed) and language of your solution. Upload working solutions into upload system as well (code in upload system must be complete with short readme how to compile and run). All your tests must be reproducible. You can be asked for reproduction of similar results during tutorial in order to prove that your solution is correct.

For testing, you can use our sever called Ritchie. You can access it by using ssh username@ritchie.ciirc.cvut.cz with CVUT password. This server is equipped with latest Intel processors running Linux and connected by 10 GBps ethernet to our test server. We also created teams channel dedicated to this server, where you can ask for installation of missing software. Unfortunately, the server is behind firewall, which blocks all outside connections except SSH. Therefore, use the hostname “ritchie.ciirc.cvut.cz” instead of public IP in server configuration in our test web application.

Fastest 12 solutions in each language will gain up to 4 extra points.

The last laboratory is reserved for presentations of the best solutions. Be ready to present your solution. :)

If you have any question, ask in the Server channel in teams.