It is hard to answer with so little information. Few notes:
- Mnesia is a very complex piece of software - there is very many
processes and ETS tables. Both of them might be a serial bottleneck.
And some serializing process/ets can be short lived.
- If you use transactions for reads and writes, you are doing
serialization of all operations, hence low CPU usage is an obvious
result of this
- If you are doing dirty operations, there might be some other locks
which are being contented. Try recompiling erlang with lock counter
enabled and taking a look at results.
On Mon, Sep 17, 2012 at 12:38 PM, Solomon <solomon.
...@gmail.com> wrote:
> I test a web server run on a 24-core server. When server get a message, it
> would read/write mnesia and return a message to client. All of the mnesia
> tables are ram_copies and disc_copies.
> I do a stress test and found that the heighest value of cpu load is only
> 900%, I get I/O load by iostat and it is idle, so I guess the system
> bottleneck is mnesia, but I found the reductions counter of mnesia's process
> are low.
> Is my guess right?
> _______________________________________________
> erlang-questions mailing list
> erlang-questi...@erlang.org
> http://erlang.org/mailman/listinfo/erlang-questions
_______________________________________________
erlang-questions mailing list
erlang-questi
...@erlang.org
http://erlang.org/mailman/listinfo/erlang-questions
