LAB #5 – mdadm Linear Device (Networking)

LAB #5 – mdadm Linear Device (Networking)
CMPSC311 - Introduction to Systems Programming
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. (S1) 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; (S2) 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; (S3) The server
wraps the fields such as opcode, return code and block (if needed) into a JBOD response message and sends
1
it to the client; (S4) the client (inside the function jbod_client_operation) next receives the response
message, extracts 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 functions will be called by the tester, not by your own code. The file net.c contains some functions with empty
bodies that can help with structuring your code, but you may implement your own help 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 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 = 0 (JBOD_MOUNT) [disk id = 0 block id = 0], result = 0
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
2
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.
Grading rubric The grading would be done according to the following rubric:
• Passing trace files with cache size 1024: 25% for simple input 30% each for random and linear trace files.
We will not measure caching efficiency and cost this time.
• Adding meaningful descriptive comments: 5%
• Successful “make” and execution without error and warnings: 5%
• Submission of commit id: 5%
Penalties: No late submission allowed. We will not grade any submissions after 04/27/2022.
3