Jared Lee Richardson [ARCHIVE] on Nostr: 📅 Original date posted:2017-03-31 📝 Original message:I guess I should caveat, a ...
📅 Original date posted:2017-03-31
📝 Original message:I guess I should caveat, a rounding error is a bit of exaggeration -
mostly because I previously assumed that it would take 14 years for
the network to reach such a level, something I didn't say and that you
might not grant me.
I don't know why paypal has multiple datacenters, but I'm guessing it
probably has a lot more to do with everything else they do -
interface, support, tax compliance, replication, redundancy - than it
does with the raw numbers of transaction volumes.
What I do know is the math, though. WW tx volume = 426,000,000,000 in
2015. Assuming tx size of ~500 bytes, that's 669 terabytes of data
per year. At a hard drive cost of 0.021 per GB, that's $36k a year or
so and declines ~14% a year.
The bandwidth is the really big cost. You are right that if this
hypothetical node also had to support historical syncing, the numbers
would probably be unmanagable. But that can be solved with a simple
checkpointing system for the vast majority of users, and nodes could
solve it by not supporting syncing / reducing peer count. With a peer
count of 25 I measured ~75 Gb/month with today's blocksize cap. That
works out to roughly 10 relays(sends+receives) per transaction
assuming all blocks were full, which was a pretty close approximation.
The bandwidth data of our 426 billion transactions per year works out
to 942 mbit/s. That's 310 Terabytes per month of bandwidth - At
today's high-volume price of 0.05 per GB, that's $18,500 a month or
$222,000 a year. Plus the $36k for storage per year brings it to
~$250k per year. Not a rounding error, but within the rough costs of
running an exchange - a team of 5 developers works out to ~$400-600k a
year, and the cost of compliance with EU and U.S. entities (including
lawyers) runs upwards of a million dollars a year. Then there's the
support department, probably ~$100-200k a year.
The reason I said a rounding error was that I assumed that it would
take until 2032 to reach that volume of transactions (Assuming
+80%/year of growth, which is our 4-year and 2-year historical average
tx/s growth). If hard drive prices decline by 14% per year, that cost
becomes $3,900 a year, and if bandwidth prices decline by 14% a year
that cost becomes $1800 a month($21,600 a year). Against a
multi-million dollar budget, even 3x that isn't a large concern,
though not, as I stated, a rounding error. My bad.
I didn't approximate for CPU usage, as I don't have any good estimates
for it, and I don't have significant reason to believe that it is a
higher cost than bandwidth, which seems to be the controlling cost
compared to adding CPU's.
> I'm not going to take the time to refute everything you've been saying
Care to respond to the math?
> This whole thread has been absurdly low quality.
Well, we agree on something at least.
On Fri, Mar 31, 2017 at 9:14 AM, David Vorick <david.vorick at gmail.com> wrote:
> No one is suggesting anything like this. The cost of running a node
> that could handle 300% of the 2015 worldwide nonbitcoin transaction
> volume today would be a rounding error for most exchanges even if
> prices didn't rise.
>
>
> Then explain why PayPal has multiple datacenters. And why Visa has multiple
> datacenters. And why the banking systems have multiple datacenters each.
>
> I'm guessing it's because you need that much juice to run a global payment
> system at the transaction volumes that they run at.
>
> Unless you have professional experience working directly with transaction
> processors handling tens of millions of financial transactions per day, I
> think we can fully discount your assessment that it would be a rounding
> error in the budget of a major exchange or Bitcoin processor to handle that
> much load. And even if it was, it wouldn't matter because it's extremely
> important to Bitcoin's security that it's everyday users are able to and are
> actively running full nodes.
>
> I'm not going to take the time to refute everything you've been saying but I
> will say that most of your comments have demonstrated a similar level of
> ignorance as the one above.
>
> This whole thread has been absurdly low quality.
📝 Original message:I guess I should caveat, a rounding error is a bit of exaggeration -
mostly because I previously assumed that it would take 14 years for
the network to reach such a level, something I didn't say and that you
might not grant me.
I don't know why paypal has multiple datacenters, but I'm guessing it
probably has a lot more to do with everything else they do -
interface, support, tax compliance, replication, redundancy - than it
does with the raw numbers of transaction volumes.
What I do know is the math, though. WW tx volume = 426,000,000,000 in
2015. Assuming tx size of ~500 bytes, that's 669 terabytes of data
per year. At a hard drive cost of 0.021 per GB, that's $36k a year or
so and declines ~14% a year.
The bandwidth is the really big cost. You are right that if this
hypothetical node also had to support historical syncing, the numbers
would probably be unmanagable. But that can be solved with a simple
checkpointing system for the vast majority of users, and nodes could
solve it by not supporting syncing / reducing peer count. With a peer
count of 25 I measured ~75 Gb/month with today's blocksize cap. That
works out to roughly 10 relays(sends+receives) per transaction
assuming all blocks were full, which was a pretty close approximation.
The bandwidth data of our 426 billion transactions per year works out
to 942 mbit/s. That's 310 Terabytes per month of bandwidth - At
today's high-volume price of 0.05 per GB, that's $18,500 a month or
$222,000 a year. Plus the $36k for storage per year brings it to
~$250k per year. Not a rounding error, but within the rough costs of
running an exchange - a team of 5 developers works out to ~$400-600k a
year, and the cost of compliance with EU and U.S. entities (including
lawyers) runs upwards of a million dollars a year. Then there's the
support department, probably ~$100-200k a year.
The reason I said a rounding error was that I assumed that it would
take until 2032 to reach that volume of transactions (Assuming
+80%/year of growth, which is our 4-year and 2-year historical average
tx/s growth). If hard drive prices decline by 14% per year, that cost
becomes $3,900 a year, and if bandwidth prices decline by 14% a year
that cost becomes $1800 a month($21,600 a year). Against a
multi-million dollar budget, even 3x that isn't a large concern,
though not, as I stated, a rounding error. My bad.
I didn't approximate for CPU usage, as I don't have any good estimates
for it, and I don't have significant reason to believe that it is a
higher cost than bandwidth, which seems to be the controlling cost
compared to adding CPU's.
> I'm not going to take the time to refute everything you've been saying
Care to respond to the math?
> This whole thread has been absurdly low quality.
Well, we agree on something at least.
On Fri, Mar 31, 2017 at 9:14 AM, David Vorick <david.vorick at gmail.com> wrote:
> No one is suggesting anything like this. The cost of running a node
> that could handle 300% of the 2015 worldwide nonbitcoin transaction
> volume today would be a rounding error for most exchanges even if
> prices didn't rise.
>
>
> Then explain why PayPal has multiple datacenters. And why Visa has multiple
> datacenters. And why the banking systems have multiple datacenters each.
>
> I'm guessing it's because you need that much juice to run a global payment
> system at the transaction volumes that they run at.
>
> Unless you have professional experience working directly with transaction
> processors handling tens of millions of financial transactions per day, I
> think we can fully discount your assessment that it would be a rounding
> error in the budget of a major exchange or Bitcoin processor to handle that
> much load. And even if it was, it wouldn't matter because it's extremely
> important to Bitcoin's security that it's everyday users are able to and are
> actively running full nodes.
>
> I'm not going to take the time to refute everything you've been saying but I
> will say that most of your comments have demonstrated a similar level of
> ignorance as the one above.
>
> This whole thread has been absurdly low quality.