Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

How to use this most efficiently? #45

Open
markg85 opened this issue Sep 14, 2022 · 3 comments
Open

How to use this most efficiently? #45

markg85 opened this issue Sep 14, 2022 · 3 comments

Comments

@markg85
Copy link

markg85 commented Sep 14, 2022

Hi,

I want to store 214 bits in a BitSet.
When i do that it's data array is 10 elements. That's 10x 32bit int (320 bit).
Ideally it would use the bare minimum in bit overhead. If that were to be done in 32 bit ints then 7 ints (224 bit) would suffice.
More ideal would be an array of chars (8 bit each) where i'd need 27 chars giving me 216 bits of room.

Are there arguments i can set in this class to get that more ideal behavior (ints or chars, both work)?

@infusion
Copy link
Collaborator

Hi
currently you can't control that behavior, since the library is optimized for speed, not memory. The 10 elements are a reasonable pre-allocation at the beginning to save re-allocations in most basic usecases < 320 bit. Especially when you narrow it down the way you're asking for you can't handle big jumps properly, such as .set(2), .set(1000). In your desired way you would increase the size linearly, instead of pre-allocating larger ranges.
Do you have a use case where 3*4 byte really make such a difference?

@markg85
Copy link
Author

markg85 commented Sep 15, 2022

Hi,

Personally i'd argue that using a bitset most often implies one is going for memory efficiency above performance.

I do have a usecase where this really matters though it's very much niche.
I'm just playing with a site to show crypto prices. I'm using the crypto.com ticker API for that which gives me that number of 214 (that many different cryptos). In my specific case i get enough data from the ticker to send an update to the browser every .05 seconds. I'm already limiting the data i send a lot to the bare minimum but that's still a list of 214 entries with the difference between the previous message.

This brings down data usage per 0.5 seconds to ~300 bytes.

Now if i have a bitset that tells me which entries changed then i only need to send:

  • that bitset (ideally 216 bits, 27 bytes)
  • each changed value (that's 20 at most in this interval thus at most 80 bytes where i assume each entry is a 4 byte float)
  • new likely max packet size ~100 bytes per 0.5 seconds

So yeah, this very niche case, does in fact save quite a bit of data if it were possible with BitSet.

@markg85
Copy link
Author

markg85 commented Sep 15, 2022

So, a very quick experiment with https://github.com/Callidon/bloom-filters/blob/master/src/bloom/bit-set.ts (from the bloom-filters package, the bitset is just for internal purposes).

That is memory efficient and it does in fact turn out that using such a structure gives me 0.5 second data blobs between 40-100 bytes. The API of that class is weird though... Oh well, works for my needs.

Could you perhaps consider adding a settable flag to your class to choose between memory efficiency and performance?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants