[erlang-questions] very large networks of erlang nodes?

Adrian Roe <>
Mon Feb 13 17:08:35 CET 2012


This was a cloud-based solution id3as built for Thomson Reuters that 
completely virtualised their online communications business that they 
run for most FTSE / NYSE companies.  Our experience was that Erlang 
played nicely when in a star topology and (understandably) got very 
upset at these levels of scale when in a fully connected environment.  
We create / throw away servers in response to demand which can be very 
spikey.

The project broke my previous "personal best" for numbers of concurrent 
servers in a single solution by about 750 ;) and I think we may be one 
of very few companies to have persuaded Amazon Cloud to return "Out of 
capacity" errors as opposed to "Exceeded account limits"...

The responsiveness of AWS is remarkable - without hot spares etc, we can 
create and commission a server entirely from scratch less than a minute 
after exceeding a load threshold.

Adrian
*Dr Adrian Roe
Director*

On 12/02/12 19:26, Miles Fidelman wrote:
> Steve Strong wrote:
>> The largest the we have run at id3as so far is 768 erlang nodes, 
>> distributed across 5 data centers. The nodes were structured in a 
>> hierarchy and used the hidden flag to ensure that the network did not 
>> become fully connected.
>>
>>
> cool! can you say a little more about what you were doing with them?
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20120213/f84162bf/attachment.html>


More information about the erlang-questions mailing list