Key Sequence

Notation

New Concepts

• permutation
• combination
• grouping

Important Results / Claims

• permutation with indistinct objects
• grouping with entirely indistinct objects (divider method)

Questions

Interesting Factoids

FS

main challenges

• naming: how do users name files
• reliability: surviving OS crashes and hardware failures
• protection: isolation between users, controlled sharing
• disk space management: minimize seeks, sharing space (“preventing fragmentation”)

seeks

to wait until the platter go under the arm and read.

internal v. external fragmentation

• internal: a file can be no less than a single block of text.
• external: no space is available even if the space in aggregate is available

main designs

contiguous allocation

IBM used this? puts files and meta-data together + implement an explicit free list allocator. benefit: simple; drawback: 1) external fragmentation 2) hard to grow files

KEY IDEAS:

1. filesystems - how do we design filesystems to manage files on disk
2. multiprocessing - how does programs interact with one another, coordinating, etc.
3. multithreading - how can we have single-process concurrency
4. virtual memory - how can one set of memory can be shared among several processes
5. modern technologies - busy waiting locking, Flash Storage, etc.

• interplay between tech + OS: OS at the hardware, software boundary
• designing with tradeoffs: not always one “best” way - evaluating pros/cons/priorities
• virtualization: make one thing look like something else, or many of them
• concurrency: synchronization is hard
• locality: predicting the future (scheduling, paging, block cache, etc.)—try to estimate the future with priority queues, etc.
• atomics: collections of operations that make them appear as a single, indivisible operation — synchronization + file system consistency (log transactions)
• layering: building higher level abstractions to hide details (monitors, fs layers, file descriptors, etc.)
• system builders wrangling complexity: solving complex problems with simple interfaces that others can build on (virtual memory, filesystems, etc.)
• trust: we have to trust something or someone—evaluating what to trust and how systems can incorporate trust

1. understanding justifies how complex systems work
2. elegant ideas of computing (concurrency, virtualization, etc.)
3. take advantage of hardware and OS software that’s available
4. OS aren’t standing still: OS changing and encountering new challenges

Massive Review

• “for optimial performance all three factors must be scaled up in tandem”
• “model performance depends strongly on scale, the model performance, the dataset size, and [compute?]”

key issues

• copyright infringement
• privacy viloating information
• information hazards reenactment

specification gaming

specification gaming, or reward hacking, is the phenomina where a system runs suboptimally because it exploited an underspecified part of the reward.

challenges

• sparse rewards
• partial obervability
• dynamic rewards (and reward shifting)
• sim-to-real transfer is hard
• computational costs
• specification gaming

AI alignment

AI alignment ensures that AI systems are aligned with human values and interests.

there is a spectrum of unexpected solutions: undesirable novel solutions an desirable novel solutions