MIME-Version: 1.0
Sender: gmaxwell@gmail.com
In-Reply-To: <CAFp6fsHwdAbcYVdPzuDdD1OV7ibCN-ebMBi113m5-JOoJtccPQ@mail.gmail.com>
References: <CADZtCShywq_s9ZK3oBBTUdCgjrTxbyeb-p7QrhJJ3mwRAECCMA@mail.gmail.com>
	<CAAS2fgROykenGH43_FXun+q4u+=94tKqnRW=kQQ2AQFeXKdW=Q@mail.gmail.com>
	<CAFp6fsHwdAbcYVdPzuDdD1OV7ibCN-ebMBi113m5-JOoJtccPQ@mail.gmail.com>
From: Gregory Maxwell <greg@xiph.org>
Date: Wed, 13 Dec 2017 00:01:32 +0000
Message-ID: <CAAS2fgRLC7GZ6pr+o59e8SycxfuBniLObFEz5yCz_fpvN8i+_Q@mail.gmail.com>
To: Suhas Daftuar <sdaftuar@gmail.com>
Content-Type: text/plain; charset="UTF-8"
Cc: Bitcoin Protocol Discussion <bitcoin-dev@lists.linuxfoundation.org>
Subject: Re: [bitcoin-dev] "Compressed" headers stream
Precedence: list

On Tue, Dec 12, 2017 at 9:07 PM, Suhas Daftuar <sdaftuar@gmail.com> wrote:
> But I think we should be able to get nearly all the benefit just by
> including nBits in any messages where the value is ambiguous; ie we include
> it with the first header in a message, and whenever it changes from the
> previous header's nBits.

Yes, that is what I was thinking last time we discussed it, just with
each header include a one byte flag that lets you express:

bit: meaning
(0) if nbits is the same as last,
(1) if timestamp is a full field or a small offset, (e.g. two bytes
defined as unsigned offset between the last time - 7200 and the new
time).
(2,3,4) if version is the same as the last distinct value .. 7th last,
or a new 32bit distinct value;
(5,6) if prev is entirely there, entirely gone, first 4 bytes
provided, or first 8 bytes provided. (if provided they override the
computed values).

That would be 7 bits in total; the 8th could be reserved or use to
signal "more headers follow" to make the encoding self-delimiting.

The downside with nbits the same as last as the optimization is that
if we ever change consensus rules to ones where difficulty management
works differently it may be the case that nbits changes every block.

Alternatively, nbits could get a differential encoding that could be
opted into for small differences-- though I haven't thought much about
it to see if a one byte difference would be that useful (e.g. can bch
differences usually be expressed with one byte?)

I'm kind of dubious of the consensus layer anti-dos separation:  nbits
minimum is so low compared to the speed of a mining device, virtually
any attack that you might do with invalid headers could still be done
with headers at the minimum difficulty. But I'm fully willing to
accept that simpler is better...


>> I would rather not change the serialization of existing messages,
>> nodes are going to have to support speaking both messages for a long
>> time, and I think we already want a different protocol flow for
>> headers fetching in any case.
>
>
> I agree with this.  Specifically the way I envisioned this working is that
> we could introduce a new 'cmpctheaders'/'getcmpcthdrs' message pair for
> syncing using this new message type, while leaving the existing
> 'headers'/'getheaders' messages unchanged.  So when communicating with
> upgraded peers, we'd never use 'getheaders' messages, and we'd only use
> 'headers' messages for potentially announcing new blocks.
>
> Of course, we'll have to support the existing protocol for a long time.  But
> one downside I've discovered from deploying BIP 130, which introduced block
> announcements via headers messages, is that overloading a 'headers' message
> to be either a block announcement or a response to a 'getheaders' message
> results in p2p-handling logic which is more complicated than it needs to be.
> So splitting off the headers chain-sync functionality to a new message pair
> seems like a nice side-effect benefit, in the long run.
>