Are High Disk Writes Expected?

[vxyz123456]

not very much RAM was being used according to Xubuntu's built-in task manager

Did you miss this part?

The other memory is not empty, it's probably in use as file cache.

and when I put in 16 GB instead of 8 GB of RAM, I noticed absolutely no difference.

No difference in memory usage, or no difference in disk writes? I guess it depends on what your Electrum server is doing: Bitcoin Core would benefit from larger dbcache, I don't know anything about Electrum server.

I highly doubt the processor would impact the amount of disk writes.

Agreed.

Both. I wasn't seeing any difference in the RAM usage or any difference in the disk writes when using 16 GB of RAM relative to 8 GB, even after increasing the dbcache in Bitcoin Core and the CACHE_MB in electrumx.

Also, Fulcrum finally fully finished indexing/processing all the blocks, so I'll let it run for a couple hours and monitor the disk writes. It will be interesting to compare to electrumx.

[larry_vw_1955]

I've never used Electrum server, but I'm curious: what could it possibly be doing that requires so much disk writes?

doesn't electrumx have a log file? why not open it and take a look at what it's doing? maybe it will tell  you when it is doing disk writes and what they are for. any reputable software would.

[ABCbits]

I've never used Electrum server, but I'm curious: what could it possibly be doing that requires so much disk writes?

doesn't electrumx have a log file? why not open it and take a look at what it's doing? maybe it will tell  you when it is doing disk writes and what they are for. any reputable software would.

Yes, it create log file. But based on previous discussion and example on https://electrumx.readthedocs.io/en/latest/HOWTO.html#understanding-the-logs, i expect bunch of these which isn't very helpful.

Code:

INFO:BlockProcessor:flushed to FS in 0.3s
INFO:BlockProcessor:flushed history in 13.4s for 934,933 addrs
INFO:BlockProcessor:flushed 6,403 blocks with 5,879,440 txs, 2,920,524 UTXO adds, 3,646,572 spends in 93.1s, committing...

But if you or someone else know how to make ElectrumX perform flush less frequently, it probably can reduce it's write activity.

[vxyz123456]

I think I'm just going to use Fulcrum. It's working well, and these results are promising. I have been running iotop for about 14 hours now (first column is cumulative disk writes, second column is process name):

Code:

506.94 M bitcoin-qt -min -chain=main [b-qt-init]
466.48 M systemd-journald
362.11 M Fulcrum /home/user/Desktop/fulcrum/fulcrum-user-config.conf [rocksdb:low]
244.21 M Fulcrum /home/user/Desktop/fulcrum/fulcrum-user-config.conf [Controller]
223.21 M bitcoin-qt -min -chain=main [b-scheduler]
212.82 M Fulcrum /home/user/Desktop/fulcrum/fulcrum-user-config.conf [rocksdb:high]
167.34 M bitcoin-qt -min -chain=main [b-msghand]
166.64 M tor --defaults-torrc /usr/share/tor/tor-service-defaults-torrc -f /etc/tor/torrc --RunAsDaemon 0
157.10 M [jbd2/sda2-8]
 94.43 M Fulcrum /home/user/Desktop/fulcrum/fulcrum-user-config.conf [rocksdb:low]
 25.96 M rsyslogd -n -iNONE [rs:main Q:Reg]
 10.98 M fwupd
  4.83 M Fulcrum /home/user/Desktop/fulcrum/fulcrum-user-config.conf
  4.27 M Fulcrum /home/user/Desktop/fulcrum/fulcrum-user-config.conf
972.00 K bitcoin-qt -min -chain=main [b-addcon]
216.00 K smartd -n
176.00 K snapd
124.00 K bitcoin-qt -min -chain=main [b-opencon]

Between all the Fulcrum variations in the list above, it looks like just under 1 GB was written in 14 hours. In the same timeframe, electrumx would have written around 70 gigabtyes to disk (seemed to average around 5 GB per hour). Crazy difference.

There are three possible causes I can think of for now:

1. I have electrumx misconfigured somehow. If other people were observing this, then this is probably not the true. However, nobody else running electrumx reported back to me, so I have no evidence that this is true.
2. Some of the code is poorly implemented somewhere or there is a bug in electrumx. I also have no evidence for this.
3. This is a difference between leveldb and rocksdb. I only tested electrumx with leveldb. Fulcrum uses rocksdb. Each of these simply may handle things differently. It's possible I wouldn't observe this if I used electrumx with rocksdb instead.

At this point, I'm not really willing to sink any more time into this to get to the root cause, since Fulcrum is working great. I'm just going to continue using it for now.

Thanks for your help, everyone.

[ABCbits]

Between all the Fulcrum variations in the list above, it looks like just under 1 GB was written in 14 hours. In the same timeframe, electrumx would have written around 70 gigabtyes to disk (seemed to average around 5 GB per hour). Crazy difference.

This is interesting discovery. Usually i would suggest people to consider ElectrumX or Fulcrum when they want to run Electrum server, but it'll make me prioritize to recommend Fulcrum instead.

There are three possible causes I can think of for now:

1. I have electrumx misconfigured somehow. If other people were observing this, then this is probably not the true. However, nobody else running electrumx reported back to me, so I have no evidence that this is true.

If you strictly follow the guide, i think it means ElectrumX have default poor configuration rather you misconfigured ElectrumX.