An evaluation of Erlang global process registries: meet Syn

#1

Post by ostinelli » Mon, Jul 06, 2015, 3:03 PM UTC

An evaluation of Erlang global process registries: meet Syn
ostinelli.net

Re: An evaluation of Erlang global process registries: meet Syn

#2

Post by rdtsc » Mon, Jul 06, 2015, 3:48 PM UTC

This looks great. Thanks for sharing. Those numbers in the end look impressive. Unlike other solutions it seems like going from 2 to 4 nodes didn't make much of an impact on regs/sec.

Re: An evaluation of Erlang global process registries: meet Syn

#3

Post by ostinelli » Mon, Jul 06, 2015, 4:13 PM UTC

rdtsc wrote: post #2

This looks great. Thanks for sharing. Those numbers in the end look impressive. Unlike other solutions it seems like going from 2 to 4 nodes didn't make much of an impact on regs/sec.

Hi, as per what I said in the article: this is just one type of use case, I'd be interested in knowing other people evaluations as well.

Re: An evaluation of Erlang global process registries: meet Syn

#4

Post by mml » Mon, Jul 06, 2015, 4:59 PM UTC

First thing I thought of was an mnesia table, glad to even be in the same ballpark as Mr. Ostinelli.

Is the slowdown as the cluster size increases is purely due to cluster gossip?

Re: An evaluation of Erlang global process registries: meet Syn

#5

Post by ostinelli » Mon, Jul 06, 2015, 5:02 PM UTC

mml wrote: post #4

First thing I thought of was an mnesia table, glad to even be in the same ballpark as Mr. Ostinelli. Is the slowdown as the cluster size increases is purely due to cluster gossip?

That's an interesting question, since Syn only uses dirty writes (and hence has no locking mechanism). Therefore yes, I think this is due to the increased nodes inter-communication (which obviously takes resources).

Re: An evaluation of Erlang global process registries: meet Syn

#6

Post by mml » Mon, Jul 06, 2015, 5:27 PM UTC

ostinelli wrote: post #5

mml wrote: post #4

First thing I thought of was an mnesia table, glad to even be in the same ballpark as Mr. Ostinelli. Is the slowdown as the cluster size increases is purely due to cluster gossip?

That's an interesting question, since Syn only uses dirty writes (and hence has no locking mechanism). Therefore yes, I think this is due to the increased nodes inter-communication (which obviously takes resources).

I recall reading at some point that things really start to get out of hand re: gossip somewhere around 12 nodes. Not sure if that's a modern limit or not though.