Re: [bitcoin-dev] Block Batch Filters for Light Clients

From: Tamas Blummer <tamas.blummer@gmail•com>
To: "admin@bitaps•com" <admin@bitaps•com>,
	Bitcoin Protocol Discussion
	<bitcoin-dev@lists•linuxfoundation.org>
Subject: Re: [bitcoin-dev] Block Batch Filters for Light Clients
Date: Sat, 21 Sep 2019 23:16:25 +0200	[thread overview]
Message-ID: <1ACE761E-A91B-45C7-A0EB-BD66FE3AD7BD@gmail.com> (raw)
In-Reply-To: <236E3AB2-4035-4F51-84EE-6F7F57298777@bitaps.com>

[-- Attachment #1: Type: text/plain, Size: 3042 bytes --]

Hi Aleksey,

Yes, BIP158 uses the block hash to seed the hash function, which makes distinct block filters non-aggregatable 
for common values. Aggregate fiters on ranges of blocks would have to use some other seed and then 
achive significant savings using the same design.

I think that the most likely use of filters is to decide if a newly announced block should be downloaded and 
not scanning over the entire chain, where aggregate filters would help. I also suspect that whole chain 
scans would be better served with plain sequential reads in map-reduce style.

Typical clients do not care of filters for blocks before the birth date of their wallet’s keys, so they skip over the 
majority of history which is a bigger saving than any aggregate filter.

I wish we get a filter committed as commitment would unlock more utility than any marginal savings through
more elaborate design.

Tamas Blummer

> On Sep 19, 2019, at 19:20, admin--- via bitcoin-dev <bitcoin-dev@lists•linuxfoundation.org> wrote:
> 
> Hello list, 
> 
> Here is a link for a draft of a BIP for  compact probabilistic block filters alternative of BIP 158
> 
> https://docs.google.com/document/d/1jH9tEUyb9w2OZd4-kxfGuyNIIZzmgkEb_z0qSxv80ik/edit?usp=sharing <https://docs.google.com/document/d/1jH9tEUyb9w2OZd4-kxfGuyNIIZzmgkEb_z0qSxv80ik/edit?usp=sharing>
> 
> Summary:
> 
>  - BIP 158  false positive rate is low, we can achieve lower bandwidth with higher false positive rate filter while sync blockchain
> 
>  - BIP 158 not do not support filter batching by design of used parameters for siphash and Golomb coding optimal parameters
> 
>  - Alternative compression with delta coding and splitting data to 2 bit string  sequences. First for data without prefixes, second one for information about  bit length written to first sequence.
>    Second sequence have a lot of duplicates,  compressed with 2 round of Huffman algorithm. (Effectivity about 98% vs Golomb with optimal parameters)
> 
>  - Block filters batching reduce filter size significantly
> 
> - Separation of filters by address type allows lite client not to download redundant information without compromising privacy.
> 
> - Lite client filters download strategy: get biggest filter (smallest blocks/size rate) for blocks range, in case positive test  -> get medium filters to reduce blocks range ->  get block filters for affected range -> download affected blocks over TOR 
> 
> Implementation (python): https://github.com/bitaps-com/pybtc/blob/bugfix/pybtc/functions/filters.py#L172 <https://github.com/bitaps-com/pybtc/blob/bugfix/pybtc/functions/filters.py#L172>
> 
> Exactly information from mainnet  about size for separated filters by address types and batch size will be added within few days.
> 
> Thanks for any feedback.
>       Aleksey Karpov
> 
> _______________________________________________
> bitcoin-dev mailing list
> bitcoin-dev@lists•linuxfoundation.org
> https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev

[-- Attachment #2: Type: text/html, Size: 4967 bytes --]