Absolutely Alex, as soon as the server goes public I'll be glad to provide a
testimonial for MINA. Additionally, we will be producing a rather
substantial set of load testing results and posting them online. They
include environment configurations, OS/hardware specs, etc. We are also
giving away the load tester source code we used for the application. It's
quite specific to our server but might help someone else out.
I will say this, generating that much load takes more machines then the
server system itself. It's also a tricky piece of code. Anyone making a load
generator should plan way ahead if they want it to work properly. We also
ran into some very strange problems. Most network developers know about file
descriptor limits but what about ephemeral ports? MINA works so well that we
had to raise the ephemeral port limit on the load generators to prevent us
from running out of outbound ports! For details:
http://www.ncftp.com/ncftpd/doc/misc/ephemeral_ports.html
Thanks!
Mike
----- Original Message -----
From: "Alex Karasulu" <[EMAIL PROTECTED]>
To: <[email protected]>
Sent: Saturday, February 24, 2007 8:39 AM
Subject: Re: MINA 1.0.2 Scalability?
Wow this is great Michael. I think this is a great candidate for our
testimonials section. WDYT?
Alex
On 2/24/07, Michael Grundvig <[EMAIL PROTECTED]> wrote:
I've been meaning to post this for a while but have been lazy. With Mr.
Royal's help, a new project my company has been working on has been able
to
achieve 120,000 concurrent connections with MINA.
We are using a distributed model that looks something like this: client
-->
"gateway" --> server where many gateways front-end the server. By
offsetting
the connection overhead to gateways, we have been able to scale much
farther
then normal. Depending on the number of messages per second, it's
possible
to scale even farther. We believe a single server could handle 1/4
million
users or more without any changes (assuming you had the appropriate
number
of gateways). In the end, garbage collector churn has been the limiting
factor in every test. Once we hit 50,000+ inbound messages per second on
the
server any garbage collector delays become a bottleneck from which we
can't
always recover. Work is being done to correct this though by reducing
object
churn.
MINA has been the single largest reason we were able to create the
application at all. Simply put, MINA works beautifully with an elegant
and
powerful design. A big thank you goes out from me to the MINA developers
and
contributors.
Thanks!
Mike
P.S. When the server goes public (it's a purchasable product) I'll go
ahead
and publish the final load test results. Right now we are working on
making
it so servers can be clustered together. This will raise the user limit
potentially into the millions. MINA rocks!
----- Original Message -----
From: "Marc Boorshtein" <[EMAIL PROTECTED]>
To: <[email protected]>; "Apache Directory Developers List"
<[email protected]>
Sent: Friday, February 23, 2007 2:26 PM
Subject: MINA 1.0.2 Scalability?
>I was curious if any scalability testing has been done on MINA based
> servers and any scalability tips that may be available. I've already
> upped my SocketAcceptor threads which is not seeming to help. MyVD
> uses MINA for it's protocol stack (based on the apacheds stack). No
> matter what I do, I can't seem to go beyond 10 concurrent connections
> and was wondering if anyone else had seen this?
>
> Thanks
> Marc
