[erlang-questions] minimizing the footprint of a running node

[Kenneth Lundin]

On Thu, Oct 30, 2008 at 8:19 PM, Joel Reymont <joelr1@REDACTED> wrote:
> I only have 3Gb of memory in my accelerator at Joyent and I'm trying
> to launch 10 nodes. beam starts with RSS of about 135Mb and quickly
> goes to 500Mb+.
>
> How can I minimize the footprint of my running nodes?
>
> I'm starting slaves like this:
>
> common_args() ->
>     "+K true +P 134217727 -smp disable".
>
You will lower the memory footprint significantly by removing +P option or
using a much lower value for it (the number of simultaneously existing
processes that
the system can handle). I am pretty sure you don't need the maximum
value there, you will run out of memory in one node before you reach
134 millions of processes and I am
pretty sure that your system will not handle millions of simultaneous
poker players (in one Erlang-node)
The value +P 134217727 will result in a memory requirement of 4 times
134217727 bytes = 536870908 bytes = 536 Mbytes for the Erlang node
just to hold the process-table.


/Kenneth Erlang/OTP, Ericsson


> start_slave_node(Name, Args) ->
>     case slave:start_link(net_adm:localhost(), Name, Args) of
>         {ok, Node} ->
>             timer:sleep(1000),
>             %%mnesia:add_table_copy(schema, Node, ram_copies),
>             rpc:call(Node, mnesia, start, []),
>             rpc:call(Node, mnesia, change_config, [extra_db_nodes,
> [node()]]),
>             timer:sleep(1000),
>             Node;
>         Reason ->
>             io:format("Failed to start slave node: ~p. Retrying in 1
> second.~n",
>                       [Reason]),
>             timer:sleep(1000),
>             start_slave_node(Name, Args)
>     end.
>
> And here's the output from top:
>
> load averages:  1.37,  2.75,  2.53;                    up 0+08:07:22
> 19:16:41
> 64 processes: 63 sleeping, 1 on cpu
> CPU states: 95.9% idle,  0.3% user,  3.8% kernel,  0.0% iowait,  0.0%
> swap
> Memory: 32G phys mem, 3899M free mem, 76G total swap, 75G free swap
>
>    PID USERNAME LWP PRI NICE  SIZE   RES STATE    TIME    CPU COMMAND
>   7710 admin      3   1    0  551M  536M sleep    0:01  0.04% beam
>   7746 admin      1  59    0 5004K 2052K cpu/5    0:00  0.01% top
>   7578 postfix    1  59    0 6728K 4624K sleep    0:00  0.00% smtpd
>   7695 admin      3   1    0  547M  534M sleep    0:01  0.00% beam
>   7665 admin      3  14    0  547M 2572K sleep    0:02  0.00% beam
>   7638 admin      3   1    0  547M  329M sleep    0:01  0.00% beam
>   7740 admin      3   1    0  551M  537M sleep    0:01  0.00% beam
>   7607 admin      3   1    0  549M  536M sleep    0:07  0.00% beam
>  19718 root      13  59    0 4100K 1532K sleep    0:02  0.00% syslogd
>   7651 admin      3  19    0  547M  534M sleep    0:01  0.00% beam
>   7680 admin      3   1    0  547M  534M sleep    0:01  0.00% beam
>  19633 root      29  59    0 8748K 2552K sleep    0:00  0.00% nscd
>   7724 admin      3   1    0  551M  537M sleep    0:01  0.00% beam
>  19737 root       1  59    0   10M 2016K sleep    0:03  0.00%
> miniserv.pl
>  19762 root       1  57    0 2760K 1176K sleep    0:00  0.00%
> authdaemond
>
>        Thanks, Joel
>
> --
> wagerlabs.com
>
>
>
>
>
> _______________________________________________
> erlang-questions mailing list
> erlang-questions@REDACTED
> http://www.erlang.org/mailman/listinfo/erlang-questions
>