On Mon, 17 May 2021 21:33:23 -0700, hongy...@gmail.com wrote:

> On Ubuntu 20.04, each time when I start the machine and enter the system,
> I find that the hard disk indicator light will be flashing for a very
> long time. As a result, during this period, the data IO performance is
> very poor.
>
> Any hints for debugging/fixing this problem?

Massive I/O at startup indicates, to me, that you have one or more massively
I/O-bound processes that begin at system start. As the kernel caches I/O
buffers in order to reduce the impact of disk reads and writes on the system
responsiveness, my guess is that the startup I/O is primarily read-only and/or
write-only (otherwise caching would reduce the physical reads to only those
that could not be satisfied from the cache).

All this points to the processes and tasks, triggered at startup, that either
/read/ a lot of data or /write/ a lot of data. So /I/ would look at your
systemd config to find
- log rotation processes,
- mandb/whatis database update processes,
- slocate database update processes, and
- desktop "semantic search" cataloguing processes
along with any obvious
- package update processes, and
- telemetry processes.

As for tuning,
0) look through your system logs for any obvious clues
1) turn off any background processes that you don't need
2) evaluate the remaining "heavy hitter" processes to see
if you can alter their startup time or process priority

HTH
--
Lew Pitcher
"In Skills, We Trust"

On Tuesday, May 18, 2021 at 11:20:15 PM UTC+8, Lew Pitcher wrote:
> On Mon, 17 May 2021 21:33:23 -0700, hongy...@gmail.com wrote:
>
> > On Ubuntu 20.04, each time when I start the machine and enter the system,
> > I find that the hard disk indicator light will be flashing for a very
> > long time. As a result, during this period, the data IO performance is
> > very poor.
> >
> > Any hints for debugging/fixing this problem?
> Massive I/O at startup indicates, to me, that you have one or more massively
> I/O-bound processes that begin at system start. As the kernel caches I/O
> buffers in order to reduce the impact of disk reads and writes on the system
> responsiveness, my guess is that the startup I/O is primarily read-only and/or
> write-only (otherwise caching would reduce the physical reads to only those
> that could not be satisfied from the cache).

How to know that?

>
> All this points to the processes and tasks, triggered at startup, that either
> /read/ a lot of data or /write/ a lot of data. So /I/ would look at your
> systemd config to find
> - log rotation processes,
> - mandb/whatis database update processes,
> - slocate database update processes, and
> - desktop "semantic search" cataloguing processes
> along with any obvious
> - package update processes, and
> - telemetry processes.

It seems that I don't have them.

>
> As for tuning,
> 0) look through your system logs for any obvious clues
> 1) turn off any background processes that you don't need
> 2) evaluate the remaining "heavy hitter" processes to see
> if you can alter their startup time or process priority
>
> HTH
> --
> Lew Pitcher
> "In Skills, We Trust"