Message-ID: <502D482A.2090609@justmoon.de>
Date: Thu, 16 Aug 2012 21:21:14 +0200
From: Stefan Thomas <moon@justmoon.de>
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64;
	rv:14.0) Gecko/20120713 Thunderbird/14.0
MIME-Version: 1.0
To: bitcoin-development@lists.sourceforge.net
References: <CA+8xBpcfxdpg-z4OQab3379amznM30Ae-Kurko0BKuySwfBy+Q@mail.gmail.com>
	<20120816175637.GA13454@vps7135.xlshosting.net>
In-Reply-To: <20120816175637.GA13454@vps7135.xlshosting.net>
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Subject: Re: [Bitcoin-development] BIP 35: add mempool message
Precedence: list

> This seems safe, although it forces other full implementations that want to
> expose protocol version 60002 (or later) to also implement this. What do they
> think about this?

BitcoinJS will implement it, it's a useful feature and there is no
reason not to support it.

Two comments from my end:

- This is just a thought, but I wouldn't mind using a new inv_type for
this, e.g. MSG_MEMTX. I could conceivably see a future where broadcast
and relay txs are stored in a very fast local cache whereas the general
mempool is stored in a slower data structure. By being able to
distinguish incoming getdata requests I can save a few milliseconds by
querying the right storage right away. Might also help with things like
telling apart broadcast/relayed transactions from the response to a
mempool request for purposes like DoS scoring etc.

Not a big deal by any means, but I also don't see a downside to it.
inv_types are not a scarce resource, we have four billion of them available.

For now clients would just treat MSG_TX and MSG_MEMTX interchangeably.

- If a node doesn't have anything in it's mempool it sends back an empty
inv message. This is either ambiguous (if other things also send empty
inv messages in the future) or arbitrary (why should an empty inv be
associated with a mempool request of all things.) Instead why not
respond with an inv message that contains a single element of type
MSG_MEMTX and hash 0. That would a very direct way to indicate that this
response is associated with a mempool request.


I'm not married to either suggestion, just trying to add my perspective.
One thing you notice when reimplementing Bitcoin is that Bitcoin's
protocol leaves out a lot of information not for space reasons, but
because the reference client's implementation doesn't happen to need it.
Sometimes however this locks other clients into doing things the same
way. If we can make the protocol a bit richer, especially if this
doesn't cost any extra bytes, then we should consider it as it might
help some implementation down the road make a neat optimization.


On 8/16/2012 7:56 PM, Pieter Wuille wrote:
> On Thu, Aug 16, 2012 at 01:32:04PM -0400, Jeff Garzik wrote:
>> Consensus was we should do a BIP for all P2P changes, so here it is...
>>  feedback requested.
>>
>> https://en.bitcoin.it/wiki/BIP_0035
> I like the idea of being able to query the memory pool of a node; the
> implementation is straightforward, which is good. Maybe effectively using the
> command can be added? I suppose it is interesting in general for nodes to
> get a memory pool refill at startup anyway.
>
>> 1) Upon receipt of a "mempool" message, the node will respond
>>    with an "inv" message containing MSG_TX hashes of all the
>>    transactions in the node's transaction memory pool.
>>
>>    An "inv" message is always returned, even if empty.
> I'm not sure about this last. What is it good for? inv packets can always be
> sent, even not in response to others, so it is not that this gives you an
> acknowledgement the mempool is updated?
>
>> 3) Feature discovery is enabled by checking two "version" message attributes:
>>
>>    a) Protocol version >= 60002
>>    b) NODE_NETWORK bit set in nServices
> This seems safe, although it forces other full implementations that want to
> expose protocol version 60002 (or later) to also implement this. What do they
> think about this?
>
> I would like to suggest to allocate an extra service bit for this. We still
> have 63 left, and this is a well-defined and useful extra service that was
> not yet provided by any earlier node. Doing that would also mean that
> mempool-providing survices may be discovered before connecting to them, as
> the service bits are carried around in addr messages. Any opinions about that?
>