At merkle we run reth in production. Syncing a new node takes a week on fast hardware:
Syncing this node took 7 days
This server has 128GB of RAM, fast NVMe disks and 1 Gbp/s bandwidth in a datacenter on the USA east coast. We have been looking for ways to speed up the process. Snapshots have been the most effective so far.
Generating a snapshot
We currently run 4 reth nodes in production. We use one of them to periodically generate a snapshot.
The process is simple. First we stop the node:
sudo systemctl stop reth
Then we compress the
mdbx.dat file (the reth database) located at
~/.local/share/reth/mainnet/db/ by default using
lz4 (best ratio for speed and compression), which takes about 1 hour on our hardware:
sudo tar --use-compress-program=lz4 -cvf /root/.local/share/reth/mainnet/db/reth.tar.lz4 /root/.local/share/reth/mainnet/db/mdbx.dat
Then, we upload the snapshot to a remote server, within our infrastructure with transfer speeds of 5 GB/s.
This setup allows us to sync a new node in less than four hours instead of 7+ days, speeding up the process 40x fold.
We've been looking for ways to share our snapshots with the community. We have looked at multiple options and finally settled on CloudFlare R2 for their unlimited egress.
As of today, in addition to our private snapshots, we are also sharing public snapshots. You can find them at snapshots.merkle.io. New snapshots are uploaded every Monday and Thursday.
In order to offer snapshots for free to everyone, we had to compromise on download speed (only ~100-300MB/s with R2). However, we can set you up with faster snapshots if you need them. If you're interested, please reach out to us at firstname.lastname@example.org.
Note: If you don't want to run reth in production, use our public unlimited RPC eth.merkle.io instead.