I've been thinking about trade-offs as "pick two of three" in the abstract, but the bookshelf example made it concrete. The insight that matters is: if you know your query patterns, you can optimize differently.
As a PM, I keep trying to build systems that work for "every case." But this article reminded me that's the wrong goal. The hash table works because it accepts the space-time trade-off. The heap works because it embraces disorder for non-priority items.
Sometimes the best system isn't the most elegant one—it's the one that matches how you'll actually use it.
Good reminder to stop over-optimizing for flexibility I'll never need.
The best way to store information depends on how you intend to use (query) it.
The query itself represents information. If you can anticipate 100% of the ways in which you intend to query the information (no surprises), I'd argue there might be an ideal way to store it.
as a line of thought, it totally does. you just extend the workload description to include writes. where this get problematic is that the ideal structure for transactional writes is nearly pessimal from a read standpoint. which is why we seem to end up doubling the write overhead - once to remember and once to optimize. or highly write-centric approach like LSM
I'd love to be clued in on more interesting architectures that either attempt to optimize both or provide a more continuous tuning knob between them
This is exactly right, and the article is clickbait junk.
Given the domain name, I was expecting something about the physics of information storage, and some interesting law of nature.
Instead, the article is a bad introduction to data structures.
One format I'm missing: storage for conversations and social media posts. Both are complex media (text + images/videos + metadata), and one is actually a collection of such posts.
How would you go about storing those in a somewhat human-readable format? My goal is to archive my chats and social media activity.
There are, however, several objectively bad ways. In "Service Model" (a novel that I recommend) a certain collection of fools decides to sort bits by whether it's a 1 or a 0, ending up with a long list of 0's followed by a long list of 1's.
Pedantic, but the article is talking about the way we structure/organize information, not store it. When I think of the word store, I think of the physical medium. The way we organize the information is only partially related
would it be more accurate to say "to store using information, using information"? Since everything ultimately boils down to information, humans trying to store information is a bit recursive?
I also like (old) .ini / TOML for small (bootstrap) config files / data exchange blobs a human might touch.
+
Re: PostgreSQL 'unfit' conversations.
I'd like some clearer examples of the desired transactions which don't fit well. After thinking about them in the background a bit I've started to suspect it might be an algorithmic / approach issue obscured by storage patterns that happen to be enabled by some other platforms which work 'at scale' supported by hardware (to a given point).
As an example of a pattern that might not perform well under PostgreSQL, something like lock-heavy multiple updates for flushing a transaction atomically. E.G. Bank Transaction Clearance like tasks. If every single double-entry booking requires it's own atomic transaction that clearly won't scale well in an ACID system. Rather the smaller grains of sand should be combined into a sandstone block / window of transactions which are processed at the same time and applied during the same overall update. The most obvious approach to this would be to switch from a no-intermediate values 'apply deduction and increment atomically' action to a versioned view of the global data state PLUS a 'pending transactions to apply' log / table (either/both can be sharded). At a given moment the transactions can be reconciled, for performance a cache for 'dirty' accounts can store the non-contested value of available balance.
I've been thinking about trade-offs as "pick two of three" in the abstract, but the bookshelf example made it concrete. The insight that matters is: if you know your query patterns, you can optimize differently.
As a PM, I keep trying to build systems that work for "every case." But this article reminded me that's the wrong goal. The hash table works because it accepts the space-time trade-off. The heap works because it embraces disorder for non-priority items.
Sometimes the best system isn't the most elegant one—it's the one that matches how you'll actually use it.
Good reminder to stop over-optimizing for flexibility I'll never need.
Thanks for sharing.
The query itself represents information. If you can anticipate 100% of the ways in which you intend to query the information (no surprises), I'd argue there might be an ideal way to store it.
I'd love to be clued in on more interesting architectures that either attempt to optimize both or provide a more continuous tuning knob between them
Given the domain name, I was expecting something about the physics of information storage, and some interesting law of nature. Instead, the article is a bad introduction to data structures.
* For lossless compression of generic data, gzip or zstd.
* For text, documentation, and information without fancy formatting, markdown, which is effectively a plain-text superset.
* For small datasets, blobs, objects, and what not, JSON.
* For larger datasets and durable storage, SQLite3.
Whenever there's text involved, use UTF-8. Whenever there's dates, use ISO8601 format (UTC timezone) or Unix timestamps.
Following these rules will keep you happy 80% of the time.
How would you go about storing those in a somewhat human-readable format? My goal is to archive my chats and social media activity.
Also, let us not confuse "relative" with "not objective". My father is objectively my father, but he is objectively not your father.
Conceptually similar to CAP, but with storage trade-offs. The idea is you can only pick 2 out of 3.
I also like (old) .ini / TOML for small (bootstrap) config files / data exchange blobs a human might touch.
+
Re: PostgreSQL 'unfit' conversations.
I'd like some clearer examples of the desired transactions which don't fit well. After thinking about them in the background a bit I've started to suspect it might be an algorithmic / approach issue obscured by storage patterns that happen to be enabled by some other platforms which work 'at scale' supported by hardware (to a given point).
As an example of a pattern that might not perform well under PostgreSQL, something like lock-heavy multiple updates for flushing a transaction atomically. E.G. Bank Transaction Clearance like tasks. If every single double-entry booking requires it's own atomic transaction that clearly won't scale well in an ACID system. Rather the smaller grains of sand should be combined into a sandstone block / window of transactions which are processed at the same time and applied during the same overall update. The most obvious approach to this would be to switch from a no-intermediate values 'apply deduction and increment atomically' action to a versioned view of the global data state PLUS a 'pending transactions to apply' log / table (either/both can be sharded). At a given moment the transactions can be reconciled, for performance a cache for 'dirty' accounts can store the non-contested value of available balance.