Was the bitcoin network victim of a spam attack ?

The amount of transactions that can be confirmed by the Bitcoin network are limited. In average, one block is mined every 10 minutes, confirming 1 MB of transactions. However, if there is more transactions than what can be confirmed by the network, the queue of transactions waiting to be confirmed starts to grow very fast. This can be observed by looking at the size of the mempool, which represents the size of the unconfirmed transactions a node has to keep in memory. In january 2017, the size of the mempool started to grow very fast, and people started to speculate why.

The first reason could be that the network was victim of a spam attack. To spam the bitcoin network, one would have to send a lot of big transactions with fees higher than average, thereby taking the place of the other legit transactions in the queue. This lead to a very slow confirmation time for the average bitcoin user, which sends transactions with fees lower than the ones used for the attack.

The second reason is that, just before the mempool started growing, a difficulty adjustment occured. In order to keep the rate of one block every 10 minutes, the difficulty is reajusted about every 10 days. The hashrate of the network has grown a lot during the last days, which led to a higher block rate, of about 16%. Thus, blocks were mined 16% faster than they should be, increasing the capacity of the network. This capacity was fully used by the users, but when the difficulty was adjusted, and we went back to the rate of 1 block every 10 minutes, the transaction rate which was not a problem for a network with 16% more capacity, was now too high, and unconfirmed transactions started to accumulate, until users reduced their use of the network. Here, I assume that the relation between the size of the mempool and the rate at which users issue transactions is reflexive. Users will stop sending transactions if they see it will take two days for them to confirm.

To get an insight of what happened, let’s first plot the amount of bytes per second that was added to the mempool, which is calculated from the transaction rate, and the average transaction size. The blue curve is an average made on a 30 minutes interval, and the green is made on a 3 hours interval. The time of the difficulty adjustment is indicated by the vertical line in red.

Bytes per second

We can see that the amount of bytes added to the mempool took off just after the difficulty change. This bad timing led to a big increase of the mempool.

An easy way to see the effect of the difficulty adjustment on this huge backlog would be to simulate how the bitcoin network would have reacted if no difficulty adjustment had happened.

Simulations

To simulate the transaction arrivals, one could simply use a poisson process with the rate being the average rate of the week of the attack. But this is not precise enough. So I took the transaction rates of the week, computed on a seven minutes interval. Then, I generated transaction arrival times with a poisson process with a different rate every 7 minutes, each rate being the corresponding rate of the network.

Each transaction size is taken from the average transaction sizes of the week, computed on a 8 minutes interval.

For the service times without difficulty change (or the time at which blocks would have been mined if the difficulty hadn’t been adjusted), I simply took the original service times of the network, but replaced the post-difficulty-change interarrival times with the pre-difficulty-change interarrival times.

I ran two simulations. One with the difficulty adjustment (blue), and one without (green). The moment when the difficulty changed is indicated by the vertical line in red.

Mempool size

We can see that the simulated mempool size with difficulty adjustment is quite close to the original mempool size. Obviously, the two simulations start to differ after the difficulty changes.

Even without the difficulty adjustment, the network would have suffered from a 25 MB backlog at its peak. This indicates that it may be a spam attack, most likely timed, which was amplified by the difficulty adjustment.

So.. spam or not spam ?

The 23rd at around 8PM (UTC), during the high bytes/second peak, this address was the output of a lot of big transactions with fees higher than average. All of this in the span of one hour. The inputs of those transactions behave strangely. Instead of sending everything at once, sometimes 8 or 9 transactions are used to send funds to the 1EPKSuFjTH3xbCK3b5ebaPbe8iUGvKvF8a address.

Edit: This process has been repeated on the 31rd, with this time the address 16FEW2CboX5hVk3FqxNJ3Wt5StySpu1TZH, leading to a new mempool increase.

Strange inputs

It seems like a lot if not all of these inputs have been involved in those kind of transactions in the past. Maybe those addresses are reused to perform attack on the bitcoin network ? This is a plot of the size of the mempool, and the vertical lines in red indicates when those inputs were active.

Mempool size relation

In conclusion, I don’t know if this is intentional. It seems like someone periodically performs very heavy operations, and has to transfer a lot of bitcoins. This time, however, those big transfers combined with the difficulty adjustment led to a huge increase of the mempool size.

Edit: This happened again on the 31rd

And this time, no difficulty adjustment occured, but the mempool still managed to reach 35MB. The transactions can be seen here. If anyone has any information on what those transfers could be, don’t hesitate to write a comment !

Written on January 28, 2017