Parity-ethereum: Upgrading from 1.11.5 to 1.11.6 → CPU usage through the roof

[c0deright]

• Which Parity version?: 1.11.6-beta
• Which operating system?: Linux (Ubuntu 16.04 LTS / Xenial, amd64)
• How installed?: dpkg -i
• Are you fully synchronized?: yes
• Which network are you connected to?: ethereum
• Did you try to restart the node?: yes
• machine used for: only parity testing, nothing else. nothing running besides parity and the usual stuff (syslog, ssh, postfix, getty, cron, ...). some rpc calls for monitoring every now and then, no external stress on the system. system is used to gather metrics like parity memory usage etc

I'm running a parity beta test machine for all kinds of things. Yesterday I upgraded the machine from 0.11.4 to 0.11.6 and now parity uses much more CPU. Last night the CPU credits for the AWS ec2 t2.medium instance went to zero because of that (t2 instances have a CPU credit system. using your CPU less than a certain limit earns you credits that you can use later for CPU heavy tasks).

I upgraded the instance to t2.large and made sure the volume is not the bottleneck (wait IO between 0 and 10% at times) and just see a much much higer CPU usage than with 1.11.4:

After testing 1.11.4 again, 1.11.5 and again 1.11.6 I can conclude that the issue is with the 1.11.6 release. Please see below.

[c0deright]

Testing 1.11.5 right now

Edit: the CPU increase is definitely in the 1.11.6 release. 1.11.5 doesn't show this. Will post graphs in an hour where you can clearly see this.

[c0deright]

1.11.5

1.11.5 doesn't show the high CPU usage constantly. It has burst of 100% CPU, yes, but it declines to 3-6% CPU usage afterwards real quick without problems and certainly stays below 30% CPU on average. Looking at cacti and top I would say it equals 1.11.4 in CPU usage.

AWS ec2 t2 CPU credits are increasing while running this version.

1.11.6

1.11.6 however is completely different. It starts with low CPU usage but after 3-5 minutes it gets in a mode where the lowest point of CPU usage visually observed in top (1s interval) over a timespan of several minutes is ~55%. Once in this mode parity CPU usage never drops below ~55% and is above 70% most of the time.

AWS ec2 t2 CPU credits are decreasing while running this version.

CPU usage/credits

Note: the graph shows CPU usage as seen from the hypervisor from 0-100%. Since the ec2 instance has 2 Cores you have to double the values seen in the graph to get the CPU usage as observed in top. 50% in the graph means 1 core is at 100%.

As you can see as soon as 1.11.6 is running CPU usage goes up (CPU credits go down) and vice-versa when 1.11.4 or 1.11.5 is running.

[c0deright]

I'm testing 1.11.7 right now and it looks good so far. Did you remove a CPU intense feature or how come 1.11.7 looks normal?

Will report back in a couple of hours.

[c0deright]

1.11.7 doesn't show this high CPU issue. In fact, it seems to have even lower CPU usage on average than 1.11.4 or 1.11.5.

Will close this issue now.