The site is now located at http://cs.iit.edu/~lee/cs450

• Describe the memory management facilities of the PDP11/40. Specifically, detail the physical/virtual address spaces, and the mapping mechanism used by the MMU.

• What is the resulting memory mapping by the end of line 0669? Does proc 0’s user structure necessarily begin at the beginning of the 6th physical page? Why or why not?

• What is the structure always to be found at virtual address 140000 in the kernel’s address space? What is the significance of this structure, and what must the kernel do with it when switching to a different process?

• What does the stack pointer point to after line 2228 in the first call to swtch, after the kernel booted?

• What data structures are associated with each process in v6? Describe the roles that they play during a process’s lifetime.

• Annotate every line of assembly code for the function _retu, from lines 0740-0749. What does each statement do, and why?

• Explain why the calls to savu and retu are found in the swtch routine at lines 2189, 2193, and 2228. You should describe the processes involved during the function’s execution.

1. Suppose that we would like to use a File Allocation Table (FAT) based filesystem to support a 4TB (4 x 10¹² byte) drive.

   • If files are to be allocated in 4KB (4096 byte) blocks (i.e., 4KB is the minimum granularity of allocation, and correspond to a single FAT entry), how large, at a minimum, must the FAT structure itself be? Justify your answer. (5 points)
   • If files are allocated in 1MB (2²⁰ byte) blocks, how large must the FAT structure be? (5 points)

2. Consider an i-node based filesystem such as UFS with the following properties:

   • Blocks are 8KB large
   • Block pointers are 64-bits in size

   • Each I-node consists of the following:

     • 4 single indirect block pointers
     • 2 double indirect block pointers
     • 1 triple indirect block pointer

   What is the maximum file size supported by the file system described above? Show your work. (10 points)

3. Assume that, in a RAID 3 system with byte level parity and a dedicated parity disk, that we have lost one of the 4 RAID-ed disk drives. Compute and fill in the column below with byte values recovered using the parity drive: (5 points)

   

4. Suppose that you’ve been asked by a database administrator to help recommend a RAID implementation to support a high throughput file system. Single drive failure must be tolerated, at a minimum. Due to limited funding, you are restricted to purchasing a maximum of 20 1TB drives, and it’s crucial that the final setup allow for at least 75% of the full capacity.

   Can you recommend a setup consisting of nested RAID levels, or a hybrid approach? Justify your answer. (5 points)

5. Consider a filesystem that supports the strategies of contiguous, linked, and index allocation. What criteria should be used in deciding which strategy is best used for a particular file? Use examples to justify your answer. (5 points)

6. Give an example of how Isolation (of the ACID properties) might be important in the context of a file system. (5 points)

7. Consider a filesystem with journaling support. Upon reboot following a crash, the following entries are discovered in the journal:

   <transaction>
     <del /home/lee/foo>
     <free 319>
     <free 12019>
   </transaction>
   <transaction>
     <del /home/lee/bar>
     <free 40
   

   Describe what must have happened before the filesystem crashed. What actions should be taken to ensure the filesystem is consistent? Is it possible to carry out any of logged transactions? (5 points)

8. Consider that the throughput of a file system implementation is closely tied to how well it allows for contiguous access to the data blocks of a file. While it is theoretically possible for both FAT and UFS filesystems to contain large amounts of fragmentation, the latter is typically seen as more “resistant” to the problem.

   What about the UFS filesystem makes it inherently easier to allocate file blocks in such a way as to lessen file fragmentation? What makes it more difficult when using FAT? (5 points)

9. Consider the following set of block address requests, arriving in the order specified, with a constant inter-arrival time of 5ms: 20, 11, 40, 19, 5, 6

   There is a single head, and there are 50 cylinders, each of which contains a single sector/block. Head movement (seek) time is linear in the number of sectors n to traverse, and can be described with the formula 5 + n; e.g., seeking across 4 cylinders requires 5+4=9ms. Access times are ignored for this problem.

   Given a current head position of 14, and that the previous block accessed was 8, compute the total seek time for each of the following head scheduling algorithms over the block requests given above.

   • SSTF (4 points)
   • LOOK (4 points)
   • C-LOOK (4 points)

Coded solutions must be properly formatted and indented.

1. Deadlock in the barrier non-solution

On page 25 of The Little Book of Semaphores (TLBoS) (in listing 3.4), a non-working solution to the barrier pattern is presented in which deadlock is likely to occur. It is, however, possible — though unlikely — that the code could be executed by multiple threads in such a way as to avoid deadlock. Describe such a path through the given barrier code for 5 threads.

2. Completing the Generalized Smokers Solution

The generalized solution for the smokers problem, as presented on page 119 of TLBoS (in listing 4.46), is incomplete — in addition to Pusher A, which is woken when tobacco is placed on the table, and the smoker holding tobacco (shown in listing 4.44), we need the code for Pushers B and C, and the smokers holding paper and matches to complete the solution.

Write up the code for Pushers B and C and the aforementioned smokers — you should adhere to the semaphore names and conventions established by the existing code.

3. The Dance Mixer Problem

At the Willowbrook Ballroom, the band is always playing a Waltz, a Tango, or a Foxtrot, and there is at all times a group of dancers (men and women) on hand, each of whom can dance two of these three dances.

Each dancer waits in a gender appropriate line for the dance he/she is capable of dancing until the corresponding music is played by the band. When the music starts playing, he/she waits to be paired up with a dancer of the opposing gender also waiting for the same dance. After they complete the dance, each dancer lines up for the next dance they are capable of dancing — the band continues to play the music until there are no couples on the dance floor. At this point the band starts to play music for the next dance, and the affair starts over again.

Implement a semaphore based solution (that avoids deadlock and race conditions) for the Dance Mixer problem.

Be sure to first list the variables used in your solution (names, types, and values), and then your coded solution. You should clearly identify each thread and what it represents (dancer, band, etc.) Comment the code to your satisfaction.

4. The Gaming Lab Problem

Bowing to popular demand, the residence hall association has invested in a “lab” consisting of 5 high-performance gaming PCs. Usage of the game lab is restricted, however, to students belonging to the Rho-Theta-Sigma (RTS), Rho-Pi-Gamma (RPG), and Phi-Pi-Sigma (FPS) fraternities.

At any given moment, the lab may be occupied by a group of no more than 5 students from a single fraternity, and as other students from the various fraternities arrive at the door they form separate lines waiting to enter the room. When the current group of students finishes their game and exits, another group of students formed from one of the lines enters. If the selected line has fewer than 5 students, all the students in that line will enter as a group, otherwise a group of 5 will be selected to enter.

Note that once a group of students enters the lab, no additional students may enter the room, regardless of the size of the group or which fraternity they belong to. It is also important to guarantee that no single fraternity is starved of the game lab — e.g., if RTS is currently using the lab and there is a mile-long line for RTS but one waiting student from RPG, the latter group will be allowed to enter.

Implement a semaphore based solution for the Gaming Lab problem.

Be sure to first list the variables used in your solution (names, types, and values), and then your coded solution. You should clearly identify each thread and what it represents (dancer, band, etc.) Comment the code to your satisfaction.

In addition, you may also find this PDP-11 Handbook a handy reference.

I strongly advise you to print this material out and bring it to class if you do not plan on purchasing the dead-tree version. If you do, it’s available on Amazon for much less than most other course textbooks.

If you like, you can also obtain the simulator via the git version control system by cloning the CS 450 course repository — do this with the following command:

git clone git://github.com/michaelee/cs450.git

Look under the sim/ps directory of the cloned repository.

For those interested in some extracurricular hacking, you can also find a PDP-11 simulator under the sim/simhv directory (courtesy of the Computer History Simulation Project). Compile it with make, and you’ll find the simulator binary at ./bin/pdp11.

Running v6 is then a matter of downloading the software kit, unpacking it, and attaching to it in the simulator and booting up: (I moved the pdp11 binary into the unpacked v6 package directory before starting, below)

foxtrot:uv6swre$ ./pdp11

PDP-11 simulator V3.7-3
sim> set cpu u18
sim> att rk0 unix0_v6_rk.dsk
sim> att rk1 unix1_v6_rk.dsk
sim> att rk2 unix2_v6_rk.dsk
sim> att rk3 unix3_v6_rk.dsk
sim> boot rk0
@unix

login: root
# ls -l
total 182
drwxr-xr-x  2 bin      1040 Jan  1  1970 bin
drwxr-xr-x  2 bin       352 Jan  1  1970 dev
drwxr-xr-x  2 bin       304 Aug 20 12:18 etc
drwxr-xr-x  2 bin       336 Jan  1  1970 lib
drwxr-xr-x 17 bin       272 Jan  1  1970 mnt
drwxr-xr-x  2 bin        32 Jan  1  1970 mnt2
-rw-rw-rw-  1 root    28472 Aug 20 12:01 rkunix
-rwxr-xr-x  1 bin     28636 Aug 20 11:38 rkunix.40
drwxrwxrwx  2 bin       144 Aug 20 12:14 tmp
-rwxr-xr-x  1 bin     28472 Aug 20 12:01 unix
drwxr-xr-x 13 bin       224 Aug 20 12:22 usr
drwxr-xr-x  2 bin        32 Jan  1  1970 usr2

You can find more instructions on the PDP-11 simulator here, and on using the v6 software package here.

Submissions are due 6/24/09 at the beginning of lecture. Internet/TV students please have your work submitted via the blackboard digital dropbox by then.

The Process Scheduling Simulator

Consider the following run setup for the scheduling simulator:

seed 5000
numprocs 10
firstarrival 0.0
interarrival constant 0.0
duration constant 50
cpuburst exponential 5
ioburst exponential 10

The following experiment configures the above run with the SJF, preemptive-SJF, and SJFA 0.5 (CPU burst duration predicted using an EMA with a configured alpha of 0.5) algorithms:

run alpharun algorithm SJF key "SJF"
run alpharun algorithm PSJF key "PSJF"
run alpharun algorithm SJFA 0.5 key "SJFA"

Create the necessary configuration files — you may need to add some of your own lines — and run the simulator with the given settings. Refer to the output for exercise 2. Exercises 3 and 4 asks you to make some modifications to the run settings.

cstin 0.5
cstout 0.5