Next level Bitcoin stress test -- June 29-30 13:00 GMT 2015

[KingAfurah]

On June 22nd, we initiated a relatively limited stress test on the Bitcoin blockchain to determine whether or not we alone could have a large impact on the ecosystem as a whole. While our initial tests merely created full blocks for a multiple hour period, transaction confirmation times remained within 6 blocks for most transactions. This test was not as successful as planned.

See here why the first stress test was not as successful as planned: CLICK

The result was roughly 12 hours of full blocks combined with increased confirmation times for many Bitcoin transactions. By selecting random transactions that were not initiated by our team, we were able to determine that many standard fee transactions were taking ~80 blocks before receiving a single confirmation.

Bitcoin is at a breaking point, yet the core developers are too wound up in petty arguments to create the required modifications for long term sustainability. If nothing is done, Bitcoin will never be anything more than a costly science project. By stress testing the system, we hope to make a clear case for the increased block size by demonstrating the simplicity of a large scale spam attack on the network.

The plan - We setup 32 Bitcoin servers that will send approximately 1 transactions per second each (up from 10 servers sending 2 transactions per second each) - Each of these transactions will be approximately 3kb in size and will each spend to 10-20 addresses - The outputs will then be combined to create large 15-30kb transactions automatically pointing back to the original Bitcoin servers.
Example: https://blockchain.info/tx/888c5ccbe3261dac4ac0ba5a64747777871b7b983e2ca1dd17e9fc8afb962519

•    Certain servers will be configured to include marginally larger than standard fees, thus guaranteeing delays from standard SPV wallets.

The target will be to generate 1mb worth of transaction data every 5 minutes. At a cost of 0.0001 per kb (as per standard fees) this stress test will cost approximately 0.1 BTC every five minutes. Another way to look at the cost is 0.1 BTC per full block that we generate. We have allocated 20 BTC for this test, and therefore will be able to single handedly fill 100 blocks, or 32 hours worth of blocks. However, we will stop pushing transactions after 24 hours at 13:00 GMT Tuesday June 30.

Predictions

•    Our above estimates are based on 1 block being mined every 10 minutes. This is standard, however deviations are likely to create temporary blips in our testing, in particular when there is an hour gap between confirmations.
•    The above calculations are based on each block being exactly 1mb and no other transactions appearing in the blocks. This is obviously unrealistic due to the fact that the average block size without our intervention is approximately 600kb. Furthermore, at least 30% of miners continue to cap blocks at 731kb. Others cap at 976kb.

Conclusions

For the sake of avoiding un-necessary calculations, lets assume that each block is 926kb in size, the average normal Bitcoin transaction volume is 600kb per block, and CoinWallet.eu will be pushing 2mb of transaction data into the network every 10 minutes. Under these conditions, the amount of transaction data being pushed to the network every 10 minutes (or every average block) will be ~2600kb. This will result in a 1674 kb backlog every 10 minutes.

By 13:00 GMT Monday June 29, the mempool of standard fee transactions will be 10mb By 24:00 GMT Monday June 29nd, the mempool of standard fee transactions will be 130mb By 13:00 GMT Tuesday June 30, the mempool of standard fee transactions will be 241mb

At this point the backlog of transactions will be approximately 241 blocks, or 1.67 days. When the average new transactions are factored into the equation, the backlog could drag on for 2-3 days. At this point, questions are raised such as whether or not this will cause a "crash landing." It is impossible to know with certainty, however we are anxiously looking forward to Monday.

[1713494964]

1713494964

[1713494964]

1713494964

[turvarya]

Your calculations were way off in the last test and you're using the same ones again??? Transaction fee is way higher than you're estimating once any mempool excess begins.

And btw KingAfurah is likely an alt of Anduck. Now that this person's identity is revealed hopefully they will be more responsible with what they do, instead of trying to spread fear throughout the community. https://bitcointalk.org/index.php?topic=1094865.msg11701281#msg11701281

Where does he estimate transaction fees?

[KingAfurah]

Your calculations were way off in the last test and you're using the same ones again??? Transaction fee is way higher than you're estimating once any mempool excess begins.

Last stress test (Monday, June 22nd) was not successful, because executing tens of thousands of transactions that correctly propagate to the network simultaneously is not as easy as we had expected. 10 servers sending out 2 transactions per second each were used for this test.

For the upcoming test, we will use 32 servers running bitcoind sending out 1 transaction each per second.
This way, we hope to lower the load on the individual server, hoping the bitcoind does not clog up and crash as easily.

[KingAfurah]

Where does he estimate transaction fees?

We sent all transactions with the standard fee of 10000 satoshis per kb. If we had sent with 20000 satoshis per kb, normal transactions would have experienced larger delays.

[yayayo]

Cute, CoinWallet.eu scammers start another PR-campaign.

There's no new insight to be gained from further stress tests. We all know that fees are too low, so all kinds of idiots can spam the blockchain.

For all readers that don't want to waste time by reading the entire announcement, I've produced a tl;dr version:

For the sake of avoiding un-necessary calculations, [...] It is impossible to know with certainty, however we are anxiously looking forward to Monday.

ya.ya.yo!

[ajareselde]

Whats in it for you ? I doubt you are doing this only for stress analysis, there has to be a larger point to this.
I found last stress test to be rather interesting, looking forward to second one on monday, i'll just make sure i don't have transactions around then 

cheers

[turvarya]

I take back what I said about him possibly being anduck, no way to know for sure so not worth starting a war over that.

Your calculations were way off in the last test and you're using the same ones again??? Transaction fee is way higher than you're estimating once any mempool excess begins.

And btw KingAfurah is likely an alt of Anduck. Now that this person's identity is revealed hopefully they will be more responsible with what they do, instead of trying to spread fear throughout the community. https://bitcointalk.org/index.php?topic=1094865.msg11701281#msg11701281

Where does he estimate transaction fees?

"The target will be to generate 1mb worth of transaction data every 5 minutes. At a cost of 0.0001 per kb (as per standard fees) this stress test will cost approximately 0.1 BTC every five minutes. Another way to look at the cost is 0.1 BTC per full block that we generate. We have allocated 20 BTC for this test, and therefore will be able to single handedly fill 100 blocks, or 32 hours worth of blocks. However, we will stop pushing transactions after 24 hours at 13:00 GMT Tuesday June 23."

Basically everyone else will send transactions with higher fees than 0.0001 kb and none of the spam will go through until later. Only people that could get fooked are those who send minutes before the spam begins.

He just says, what transaction fee he is using, not what someone else will use.
You predict, that the transaction fees go up.  We will see, if that is true. But looking at my Mycelium Wallet, I don't think it reacts to how many transactions are already in the mempool.

[KingAfurah]

1. You'd have to raise your transaction fees to well above .001 BTC/kb to cause any noticeable blockade, here is the last test where legit users broke through your blockade easily. And of course you couldn't afford a significant amount of spam at this level.


2. It's an interesting idea for an attack, but this is why it's never going to be feasible. Based on what I know now I honestly don't give a crap if you do this attack at full force cause it will have barely any impact for people who are properly calculating transaction fees.

1. We don't want to cause a noticable blockade*, this is why we are using standard fee (0.0001 per kb). This is a stress test, not an attack.

2. OK, so move on with your life then.

*During the stress test we found the following behaviour:

    Transactions with a fee of 1,000 satoshi (0.000 01 BTC) per KB of transaction took 87 blocks to confirm. These transactions had fees lower than the stress test transactions and so were effectively put at the back of the queue
    Transactions with a fee of 3,000 satoshi per KB took 11 or 80 blocks to confirm
    Transactions with a fee of 10,000 satoshi per KB took an average of 9 blocks to get mined

https://multibit.org/blog/2015/06/23/bitcoin-network-stress-test.html

[turvarya]

Your calculations were way off in the last test and you're using the same ones again??? Transaction fee is way higher than you're estimating once any mempool excess begins.

Last stress test (Monday, June 22nd) was not successful, because executing tens of thousands of transactions that correctly propagate to the network simultaneously is not as easy as we had expected. 10 servers sending out 2 transactions per second each were used for this test.

For the upcoming test, we will use 32 servers running bitcoind sending out 1 transaction each per second.
This way, we hope to lower the load on the individual server, hoping the bitcoind does not clog up and crash as easily.

You'd have to raise your transaction fees to well above .001 BTC/kb to cause any noticeable blockade, here is the last test where legit users broke through your blockade easily. And of course you couldn't afford a significant amount of spam at this level.



It's an interesting idea for an attack, but this is why it's never going to be feasible. Based on what I know now I honestly don't give a crap if you do this attack at full force cause it will have barely any impact for people who are properly calculating transaction fees.

Where did you get this graph from?
Could you show it in a bigger time frame? It doesn't show, if such spikes are unusual.

Looking at this graph, I don't see anything unusual for the time of the stress test.:
https://blockchain.info/charts/transaction-fees
but that, isn't a good graph either.
I'd really love a site with better statistics, if somebody knows any.

[KingAfurah]

So what's the point of this now that you know it won't actually delay legit transactions?

The point is to study the effects of a mempool size of >150MB which was never reached so far.

[Argwai96]

Is pretty interesting to watch the blockhain action when the stress test begins and has is in progress, I'll be watching this one from the beginning.

[KingAfurah]

Fair enough, I look forward to the results. Maybe you should edit your conclusions since right now it sounds like you're saying transactions will be delayed for over 200 blocks, which isnt true for most transactions. Will be business as usual for anyone that sends with a higher fee.

No, all I am saying is "backlog of transactions will be approximately 241 blocks".

High fee transaction will be business as usual for sure.

[KingAfurah]

I think the community would be more open to the test if you explicitly stated in your announcements that this won't delay transactions which use higher fees, so people won't worry about whether they can send Bitcoin during the test.

This will certainly delay normal transactions. Results from last test show that 0.0001BTC per kb transaction on average took 9 blocks (~90 minutes) to confirm.

For quick confirmations a higher fee is the option.

[yayayo]

I think the community would be more open to the test if you explicitly stated in your announcements that this won't delay transactions which use higher fees, so people won't worry about whether they can send Bitcoin during the test.

I would be more open to the test, if in addition to announcing that higher fees are necessary he would reimburse those increased fees to everyone affected.

Imho the only reasons for doing another meaningless "stress test" on the public Bitcoin blockchain (and not on testnet) are attracting public attention for own services offered and/or manipulating market prices. Ask yourselves - why is he announcing the stress test at all? Doesn't that distort the results by people adapting in advance to increased transaction fees?

ya.ya.yo!

[turvarya]

Where did you get this graph from?
Could you show it in a bigger time frame? It doesn't show, if such spikes are unusual.

Looking at this graph, I don't see anything unusual for the time of the stress test.:
https://blockchain.info/charts/transaction-fees
but that, isn't a good graph either.
I'd really love a site with better statistics, if somebody knows any.

I made it using data from http://btc.blockr.io/ with the help of google sheets https://docs.google.com/spreadsheets/d/1LmBk8_5BqFVvMDLJ2EdhUvkyA_Nea4ZFKCE4jA-VsuA/edit?usp=sharing

Would be easy to make your own table and calculate to verify.

Thanks.
I was hoping for a more sophisticated tool. It is really a pain in the ass, to make a depper analysis this way.
Maybe I can hack something together. Still if somebody knows a better tool, that is already written, I would be very thankful for that.

[KingAfurah]

2 hours and 8 minutes to go.

All 32 server are ready and connected.

[ragi]

I am just gonna ask why again?

[ChetnotAtkins]

Looking forward to the results. Bring it coinwallet.eu!

[elrippo]

Everybody fasten your seat belts and get your popcorn yeeeaaahhhh 

[primer-]

Can't wait..