LAB #5 – mdadm Linear Device (Networking)
CMPSC311 - Introduction to Systems Programming
Spring 2024 - Due date: April 24, 2024 (11:59 PM) EST
Adding a cache to your mdadm system has significantly improved its latency and reduced the load on the JBOD. Before you finish your internship, however, the company wants you to add networking support to your mdadm implementation to increase the flexibility of their system. The JBOD systems purchased by the company can accept JBOD operations over a network using a proprietary networking protocol. Specifically, a JBOD system has an embedded server component that can be configured to have an IP address and listen for JBOD operations on a specific port. In this final step of your assignment, you are going to implement a client component of this protocol that will connect to the JBOD server and execute JBOD operations over the network. As the company scales, they plan to add multiple JBOD systems to their data center. Having networking support in mdadm will allow the company to avoid downtime in case a JBOD system malfunctions, by switching to another JBOD system on the fly.
Currently, your mdadm code has multiple calls to jbod_operation, which issue JBOD commands to a locally attached JBOD system. In your new implementation, you will replace all calls to jbod_operation with jbod_client_operation, which will send JBOD commands over a network to a JBOD server that can be anywhere on the Internet (but will most probably be in the data center of the company). You will also implement several support functions that will take care of connecting/disconnecting to/from the JBOD server.
Protocol
The protocol defined by the JBOD vendor has two messages. The JBOD request message is sent from your client program to the JBOD server and contains an opcode and a buffer when needed (e.g., when your client needs to write a block of data to the server side jbod system). The JBOD response message is sent from the JBOD server to your client program and contains an opcode and a buffer when needed (e.g., when your client needs to read a block of data from the server side jbod system). Both messages use the same format:
Bytes |
Field |
Description |
0-1 |
length |
The size of the packet in bytes |
2-5 |
opcode |
The opcode for the JBOD operation (format defined in Lab 2 README) |
6-7 |
return code |
Return code from the JBOD operation (i.e., returns 0 on success and -1 on failure) |
8-263 |
block |
Where needed, a block of size JBOD BLOCK SIZE |
Table 1: JBOD procotol packet format
In a nutshell, there are four steps:
1. The client side (inside the function jbod_client_operation) wraps all the parameters of a jbod operation into a JBOD request message and sends it as a packet to the server side
2. The server receives the request message, extract the relevant fields (e.g., opcode, block if needed), issues the jbod_operation function to its local jbod system and receives the return code
3. The server wraps the fields such as opcode, return code and block (if needed) into a JBOD response message and sends it to the client
4. The client (inside the function jbod_client_operation) next receives the response message, ex- tracts the relevant fields from it, and returns the return code and fill the parameter “block” if needed.
Note that the first three fields (i.e., length, opcode and return code) of JBOD protocol messages can be considered as packet header, with the size HEADER LEN predefined in net.h. The block field can be considered as the optional payload. You can set the length field accordingly in the protocol messages to help the server infer whether a payload exists (the server side implementation follows the same logic).
Implementation
In addition to replacing all jbod_operation calls in mdadm.c with jbod_client_operation, you will implement functions defined in net.h in the provided net.c file. Specifically, you will implement jbod_connect function, which will connect to JBOD SERVER at port JBOD PORT, both defined in net.h , and jbod_disconnect function, which will close the connection to the JBOD server. Both of these func- tions will be called by the tester, not by your own code. The file net.c contains some other functions with empty bodies that can help with structuring your code, but you may implement your own helper functions as long as you implement those functions in net.h that will be directly called by tester.c and mdadm.c. That being said, following the structure would probably be the easiest way to debug/test/finish this project. Please refer to net.c for the detailed description on the purpose, parameters, and return value of each function.
Testing
Once you finish implementing your code, you can test it by running the provided jbod_server in one terminal, which implements the server component of the protocol, and running the tester with the workload file in another terminal. Below is a sample session from the server and the client:
Output from the jbod_server terminal:
$ ./jbod_server
JBOD server listening on port 3333 . . .
new client connection from 127.0.0.1 port 32402
client closed connection
Output from the tester terminal:
$ ./tester -w traces/random-input -s 1024 >x
Cost: 17669400
Hit rate: 24.5%
$ diff x traces/random-expected-output
$
You can also run the jbod_server in verbose mode (by using -v) to print out every command that it receives from the client. Below is sample output that was trimmed to fit the space.
$ ./jbod_server -v
JBOD server listening on port 3333 . . .
new client connection from 127.0.0.1 port 38546
received cmd id = 2 (JBOD_SEEK_TO_DISK) [disk id = 0 block id = 0], result = 0
received cmd id = 5 (JBOD_WRITE_BLOCK) [disk id = 0 block id = 0], result = 0 block contents:
0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00
If your implementation is correct, your output x will be the same as the expected output from each trace workload. You will use the diff command to measure the difference, as you did for lab 3 and lab 4. Since your lab 5 code does not change the caching policy, it should produce the same result as that from your lab 4. We will only consider your net.h, net.c, cache.h, cache.c, mdadm.h and mdadm.c from your submission in our test.
Deliverables: Push your code to GitHub. Submit the commit ID of your latest code to Canvas for grading. Ensure that there is no additional text, words, or comments around the commit ID. Check canvas assignment page for more details.
Grading rubric The grading would be done according to the following rubric:
• Passing trace files with cache size 1024: 25% for simple input, 35% for random and 35% for linear trace files. We will not measure caching efficiency (hit rate) and cost (even if cost is 0). We do require that cache be used correctly.
• Adding meaningful descriptive comments: 5%
• If you have any make errors or warnings or program is stuck in a loop then you will receive a straight 0. Make sure your code does not have any make error before submitting
• If you do not submit your commit ID on canvas you will receive a straight 0
Penalties: No late submission allowed. We will not grade any submissions after 04/24/2024.
版权所有:编程辅导网 2021 All Rights Reserved 联系方式:QQ:99515681 微信:codinghelp 电子信箱:99515681@qq.com
免责声明:本站部分内容从网络整理而来,只供参考!如有版权问题可联系本站删除。