Linux Network Load Balancing is out!

Hi all!
I've just released my first beta of Linux Network Load Balancing....
it's a driver (+ userland tool) to make decentered load balancing
clusters.

I hope someone is interested in the project and could do some testing
in decent environments (I've just done some "laboratory" test in a 3
virtualized node cluster environment, but nothing more than this).

Sources and some documentation on
http://lnlb.sourceforge.net/

Greetings,
Primiano Tucci



p.tucci

p.tucci <a t> gmail.com wrote:
> Hi all!
> I've just released my first beta of Linux Network Load Balancing....
> it's a driver (+ userland tool) to make decentered load balancing
> clusters.
>
> I hope someone is interested in the project and could do some testing
> in decent environments (I've just done some "laboratory" test in a 3
> virtualized node cluster environment, but nothing more than this).
>
> Sources and some documentation on
> http://lnlb.sourceforge.net/

I didn't see anything in the docs as to whether your package supports
"sticky" connections. If a web page requires session variables (for
login, settings, etc), the user needs to hit the same server every time.
Without a sticky option, this is not possible.

Yes, I know you can avoid this problem by using a shared mount to hold
sessions, but that becomes a single point of failure, which is not a
good thing on a high availability website.

--
Ogre

Ogre wrote:
> p.tucci <a t> gmail.com wrote:
>> Hi all!
>> I've just released my first beta of Linux Network Load Balancing....
>> it's a driver (+ userland tool) to make decentered load balancing
>> clusters.
>>
>> I hope someone is interested in the project and could do some testing
>> in decent environments (I've just done some "laboratory" test in a 3
>> virtualized node cluster environment, but nothing more than this).
>>
>> Sources and some documentation on
>> http://lnlb.sourceforge.net/
>
> I didn't see anything in the docs as to whether your package supports
> "sticky" connections. If a web page requires session variables (for
> login, settings, etc), the user needs to hit the same server every time.
> Without a sticky option, this is not possible.
>
> Yes, I know you can avoid this problem by using a shared mount to hold
> sessions, but that becomes a single point of failure, which is not a
> good thing on a high availability website.
>

Of course, right after posting the above I see in the docs where it says
a client is always sent back to the same node. So it is "sticky" by
default.


--
Ogre

On Tue, 25 Sep 2007 13:29:01 -0700, Ogre <(E-Mail Removed)> wrote:
> p.tucci <a t> gmail.com wrote:
>> Hi all!
>> I've just released my first beta of Linux Network Load Balancing....
>> it's a driver (+ userland tool) to make decentered load balancing
>> clusters.
>>
>> I hope someone is interested in the project and could do some testing
>> in decent environments (I've just done some "laboratory" test in a 3
>> virtualized node cluster environment, but nothing more than this).
>>
>> Sources and some documentation on
>> http://lnlb.sourceforge.net/
>
> I didn't see anything in the docs as to whether your package supports
> "sticky" connections. If a web page requires session variables (for
> login, settings, etc), the user needs to hit the same server every time.
> Without a sticky option, this is not possible.
>
> Yes, I know you can avoid this problem by using a shared mount to hold
> sessions, but that becomes a single point of failure, which is not a
> good thing on a high availability website.
>

most normal people hold sessions in mysql.

i

Ignoramus30458 wrote:
> On Tue, 25 Sep 2007 13:29:01 -0700, Ogre <(E-Mail Removed)> wrote:
>> p.tucci <a t> gmail.com wrote:
>>> Hi all!
>>> I've just released my first beta of Linux Network Load Balancing....
>>> it's a driver (+ userland tool) to make decentered load balancing
>>> clusters.
>>>
>>> I hope someone is interested in the project and could do some testing
>>> in decent environments (I've just done some "laboratory" test in a 3
>>> virtualized node cluster environment, but nothing more than this).
>>>
>>> Sources and some documentation on
>>> http://lnlb.sourceforge.net/
>> I didn't see anything in the docs as to whether your package supports
>> "sticky" connections. If a web page requires session variables (for
>> login, settings, etc), the user needs to hit the same server every time.
>> Without a sticky option, this is not possible.
>>
>> Yes, I know you can avoid this problem by using a shared mount to hold
>> sessions, but that becomes a single point of failure, which is not a
>> good thing on a high availability website.
>>
>
> most normal people hold sessions in mysql.

Not all packages support sql based session info. I have to support a
wide variety of crap, so don't have the luxury of being picky.

--
Ogre

On Tue, 25 Sep 2007 13:58:29 -0700, Ogre <(E-Mail Removed)> wrote:
> Ignoramus30458 wrote:
>> On Tue, 25 Sep 2007 13:29:01 -0700, Ogre <(E-Mail Removed)> wrote:
>>> p.tucci <a t> gmail.com wrote:
>>>> Hi all!
>>>> I've just released my first beta of Linux Network Load Balancing....
>>>> it's a driver (+ userland tool) to make decentered load balancing
>>>> clusters.
>>>>
>>>> I hope someone is interested in the project and could do some testing
>>>> in decent environments (I've just done some "laboratory" test in a 3
>>>> virtualized node cluster environment, but nothing more than this).
>>>>
>>>> Sources and some documentation on
>>>> http://lnlb.sourceforge.net/
>>> I didn't see anything in the docs as to whether your package supports
>>> "sticky" connections. If a web page requires session variables (for
>>> login, settings, etc), the user needs to hit the same server every time.
>>> Without a sticky option, this is not possible.
>>>
>>> Yes, I know you can avoid this problem by using a shared mount to hold
>>> sessions, but that becomes a single point of failure, which is not a
>>> good thing on a high availability website.
>>>
>>
>> most normal people hold sessions in mysql.
>
> Not all packages support sql based session info. I have to support a
> wide variety of crap, so don't have the luxury of being picky.
>

But what happens whan you bounce your webserver, do you lose all
sessions?

My site holds sessions up to half a year. They get cleaned out every 6
months, IIRC.

i

Ignoramus30458 wrote:
> On Tue, 25 Sep 2007 13:58:29 -0700, Ogre <(E-Mail Removed)> wrote:
>> Ignoramus30458 wrote:
>>> On Tue, 25 Sep 2007 13:29:01 -0700, Ogre <(E-Mail Removed)> wrote:
>>>> p.tucci <a t> gmail.com wrote:
>>>>> Hi all!
>>>>> I've just released my first beta of Linux Network Load Balancing....
>>>>> it's a driver (+ userland tool) to make decentered load balancing
>>>>> clusters.
>>>>>
>>>>> I hope someone is interested in the project and could do some testing
>>>>> in decent environments (I've just done some "laboratory" test in a 3
>>>>> virtualized node cluster environment, but nothing more than this).
>>>>>
>>>>> Sources and some documentation on
>>>>> http://lnlb.sourceforge.net/
>>>> I didn't see anything in the docs as to whether your package supports
>>>> "sticky" connections. If a web page requires session variables (for
>>>> login, settings, etc), the user needs to hit the same server every time.
>>>> Without a sticky option, this is not possible.
>>>>
>>>> Yes, I know you can avoid this problem by using a shared mount to hold
>>>> sessions, but that becomes a single point of failure, which is not a
>>>> good thing on a high availability website.
>>>>
>>> most normal people hold sessions in mysql.
>> Not all packages support sql based session info. I have to support a
>> wide variety of crap, so don't have the luxury of being picky.
>>
>
> But what happens whan you bounce your webserver, do you lose all
> sessions?
>
> My site holds sessions up to half a year. They get cleaned out every 6
> months, IIRC.

I'd say that's way too long. If someone hasn't logged into the system
in a month, I say let their session expire. As long as they keep coming
back at least once a month, it won't expire.

--
Ogre