[erlang-questions] mnesia feasibility

[Jesper Louis Andersen]

On Wed, Sep 3, 2014 at 12:55 PM, Madhukar Sah <madhukars@REDACTED> wrote:

> I am running 16  nodes that are constantly pushing data in mnesia database
> located in a card powered by erlang. The card is running at 260 MHz with
> 128 MB ram. The data is being maintained as RAM_COPY as well as DISC_COPY
> in the card. Each of the 16 nodes running on a PC are pumping records every
> 1 sec. The result is mnesia is overloading and after some time mnesia is
> taking around 10 seconds to write one record into mnesia table. I have
> checked the same thing by putting the data as RAM_COPY as well but mnesia
> is taking again around 10 seconds to write one record in mnesia table.


Something does not make sense here. disc_copies also stores the table in
RAM but backs it on disc, so you don't need ram_copies in that situation.
The overload messages vary, but the most common one is when mnesia is not
able to dump the in-memory table to disk before it has to dump again. You
say you have a 260mhz CPU and 128 MB ram, which is a fairly small system.
What kind of disc I/O are you getting in such a system? Some napkin math
usually helps here to figure out the location of an eventual bottleneck.



-- 
J.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20140903/5ca5a980/attachment.htm>