MIME-Version: 1.0
From: Cserveny Tamas <cserveny.tamas+bc@gmail.com>
Date: Fri, 01 Sep 2017 12:47:08 +0000
Message-ID: <CACY+MSO8PM89VTtKfuZiQGvAs7a7R6_4mY+onhZh9KnvwxW2uw@mail.gmail.com>
To: bitcoin-dev@lists.linuxfoundation.org
Content-Type: multipart/alternative; boundary="001a114e5e384f543f05582029c1"
Subject: [bitcoin-dev] Horizontal scaling of blockchain
Precedence: list

--001a114e5e384f543f05582029c1
Content-Type: text/plain; charset="UTF-8"

Hi,

I was thinking about how to scale the block-chain.

The fundamental problem is that if the miners add capacity it will only
increase (protect) their share of block reward, but does not increase the
speed of transactions. This will only raise the fee in the long run with
the block reward decreasing.
The throughput is limited by the block size and the complexity. Changing
any of variables in the above equation was raised already many times and
there was no consensus on them.

The current chain is effectively single threaded. If we look around in the
sw industry how single threaded applications could be scaled, one viable
option emerge: horizontal scaling. This is an option if the problem can be
partitioned, and transactions in partitions could be processed alongside.
Number of partitions would be start with a fairly low number, something
between 2-10, but nothing is against setting it to a higher number later on
according to a schedule.

Partitioning key alternatives:
*Ordering on inputs:*

1) In this case transactions would need to be mined per input address
partition.
2) TX have inputs in partition 1 and 2, then needs a confirmation in both
partitions.
3) All partitioned chains have the same longest valid approach.
4) Only one chain needed to be considered for double spending, others are
invalid in case they contain that input.

This opens up questions like:
- how the fee will be shared? Fees per partition?
- Ensure a good hash function which spreads evenly, because the inputs
cannot be manipulated for load balancing
- What to do about half mined transactions (Maybe they should be two
transactions and then there is less effect about it, but payment won't be
atomic in both partitions)

*Ordering on transaction ids:*

1) Transactions would be partitioned by their TX-id. Maybe a field allowing
txid-s to match a given partition.
2) Creating blocks like this parallel would be safe for bonefide
transactions. A block will be created each 10 mins.
3) In order to get malicious/doublespent transactions out of the system
another layer must be introduced.
- This layer would be used to merge the parallel blocks. It would have to
refer all previous blocks considered for unspent inputs.
- Most of the blocks will merge just fine as normally block 1 and block 2
would not contain double spending. (of course malicious double spending
will revert speed to current levels, because the miner might have to drop a
block in the partition because it contains a spent input on another
stronger branch)
- The standard longest chain wins strategy would be used for validity on
the meta level
- Meta does not require mining, a branches can be added and they are valid
unless there are double spent inputs inside. Block inside this meta already
"paid for".

Generally both ways would have an effect on the block reward and
complexity, which is needs to be adjusted. (not to create more BTC at the
end, reduced hashing power on partitions.)
I think complexity is not an issue, the important thing is that we tune it
to 10mins / block rate per partition.

Activation could be done by creating the infrastructure first and using
only one partitions only, which is effectively the same as today. Then
activate partitions on a certain block according to a schedule. From that
block, partition enforcement will be active and the transactions will be
sorted to the right partition / chain.

It is easy to make new partitions, just need to activate them on branch
block number.
Closing partitions is a bit more complex in case of TX partitioned
transactions, but managed by the meta layer and activated at a certain
partition block. Maybe it is not even possible in case of input partitions.

I could imagine that it is too big change. Many cons and pros on partition
keys.

What is your opinion about it?

Cheers,

Tamas

--001a114e5e384f543f05582029c1
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr">Hi,<div><br></div><div>I was thinking about how to scale t=
he block-chain.</div><div><br></div><div>The fundamental problem is that if=
 the miners add capacity it will only increase (protect) their share of blo=
ck reward, but does not increase the speed of transactions. This will only =
raise the fee in the long run with the block reward decreasing.=C2=A0</div>=
<div>The throughput is limited by the block size and the complexity. Changi=
ng any of variables in the above equation was raised already many times and=
 there was no consensus on them.=C2=A0</div><div><br></div><div>The current=
 chain is effectively single threaded. If we look around in the sw industry=
 how single threaded applications could be scaled, one viable option emerge=
: horizontal scaling. This is an option if the problem can be partitioned, =
and transactions in partitions could be processed alongside.=C2=A0</div><di=
v>Number of partitions would be start with a fairly low number, something b=
etween 2-10, but nothing is against setting it to a higher number later on =
according to a schedule.</div><div><br></div><div>Partitioning key alternat=
ives:</div><div><b>Ordering on inputs:</b>=C2=A0</div><div><br></div><div>1=
) In this case transactions would need to be mined per input address partit=
ion.=C2=A0</div><div>2) TX have inputs in partition 1 and 2, then needs a c=
onfirmation in both partitions.=C2=A0<br></div><div>3) All partitioned chai=
ns have the same longest valid approach.</div><div>4) Only one chain needed=
 to be considered for double spending, others are invalid in case they cont=
ain that input.</div><div><br></div><div>This opens up questions like:<br><=
/div><div>- how the fee will be shared? Fees per partition?</div><div>- Ens=
ure a good hash function which spreads evenly, because the inputs cannot be=
 manipulated for load balancing</div><div>- What to do about half mined tra=
nsactions (Maybe they should be two transactions and then there is less eff=
ect about it, but payment won&#39;t be atomic in both partitions)</div><div=
><br></div><div><b>Ordering on transaction ids:</b></div><div><br></div><di=
v>1) Transactions would be partitioned by their TX-id. Maybe a field allowi=
ng txid-s to match a given partition.<br></div><div>2) Creating blocks like=
 this parallel would be safe for bonefide transactions. A block will be cre=
ated each 10 mins.<br></div><div>3) In order to get malicious/doublespent t=
ransactions out of the system another layer must be introduced.=C2=A0</div>=
<div>- This layer would be used to merge the parallel blocks. It would have=
 to refer all previous blocks considered for unspent inputs.</div><div>- Mo=
st of the blocks will merge just fine as normally block 1 and block 2 would=
 not contain double spending. (of course malicious double spending will rev=
ert speed to current levels, because the miner might have to drop a block i=
n the partition because it contains a spent input on another stronger branc=
h)</div><div>- The standard longest chain wins strategy would be used for v=
alidity on the meta level</div><div>- Meta does not require mining, a branc=
hes can be added and they are valid unless there are double spent inputs in=
side. Block inside this meta already &quot;paid for&quot;.</div><div><br></=
div><div>Generally both ways would have an effect on the block reward and c=
omplexity, which is needs to be adjusted. (not to create more BTC at the en=
d, reduced hashing power on partitions.)</div><div><div>I think complexity =
is not an issue, the important thing is that we tune it to 10mins / block r=
ate per partition.=C2=A0</div><div><br></div>Activation could be done by cr=
eating the infrastructure first and using only one partitions only, which i=
s effectively the same as today. Then activate partitions on a certain bloc=
k according to a schedule. From that block, partition enforcement will be a=
ctive and the transactions will be sorted to the right partition / chain.<b=
r class=3D"inbox-inbox-Apple-interchange-newline"></div><div><br></div><div=
>It is easy to make new partitions, just need to activate them on branch bl=
ock number.=C2=A0</div><div>Closing partitions is a bit more complex in cas=
e of TX partitioned transactions, but managed by the meta layer and activat=
ed at a certain partition block. Maybe it is not even possible in case of i=
nput partitions.</div><div><br></div><div>I could imagine that it is too bi=
g change. Many cons and pros on partition keys.</div><div><br></div><div>Wh=
at is your opinion about it?</div><div><br></div><div>Cheers,</div><div><br=
></div><div>Tamas</div></div>

--001a114e5e384f543f05582029c1--