You are viewing a plain text version of this content. The canonical link for it is here.
Posted to modperl@perl.apache.org by Ken Perl <ke...@gmail.com> on 2006/05/09 10:47:14 UTC
clusters
Is it possible to make a modperl application to run in cluster? if
yes, how to do that? any doc?
--
perl -e 'print unpack(u,"62V5N\"FME;G\!E<FQ`9VUA:6PN8V]M\"\@``
")'
Re: clusters
Posted by David Nicol <da...@gmail.com>.
On 5/9/06, Harmen <ha...@tty.nl> wrote:
> On Tue, May 09, 2006 at 04:47:14PM +0800, Ken Perl wrote:
> > Is it possible to make a modperl application to run in cluster? if
> > yes, how to do that? any doc?
>
> Take a few mod_perl servers and put one pound (http://www.apsis.ch/pound/) reverse
> proxy in front of them.
> Works great.
A decade ago the standard answer was "run squid" but I guess I missed a meeting.
Looking at the pound web page I see that it isn't going to accelerate the static
pages, which squid will do for you.
If Ken Perl (wonderful name -- "know perl") has CPU-intensive mod-perl
applications
that he wants to use clustering to mitigate bottlenecks, instead of
load-balancing,
a loadbalancing front-end might not be the right thing. As long as
the session management
is shared among the servers, load-balancing will work (which has
alkready been said)
if you go really simple and have your cluster all with different
external IP addresses and
use a round-robin DNS, you don't have to worry about session
management sharing as
much because any client will establish an affinity for the server they
are using and won't
switch, in general.
--
David L Nicol
route around stupidity like any other damage
Re: clusters
Posted by "Kevin A. McGrail" <km...@pccc.com>.
I'm intrigued by Pound as I've never heard of it. I have been looking
heavily at using Apache 2.2 for this same purpose but pound seems elegantly
simple in it's approach. The only drawback I can see is that the lack of
caching will cause dial-up users to stay connected to the backend servers
longer than necessary.
Anyone run pound in a live environment and have some success/failure
stories? I'd especially like to know how many unique users someone might
have been able to handle with pound on what hardware.
Regards,
KAM
> backend webserver (mod_perl1, mod_perl2, plain apache, mod_ruby,
whatever). You
> can configure pound to let specific backend servers handle specific
> domains, or urls, or a combination.
Re: clusters
Posted by Harmen <ha...@tty.nl>.
On Tue, May 09, 2006 at 05:49:43PM +0800, Ken Perl wrote:
> Do you mean that the pound work great with apache 2.0 and apache 1.3 both?
It doesn't matter.
The clients talk to the pound server. Pound proxies every request to a
backend webserver (mod_perl1, mod_perl2, plain apache, mod_ruby, whatever). You
can configure pound to let specific backend servers handle specific
domains, or urls, or a combination.
The trick is to let multiple backend servers handle requests for a domain.
Pound will notice if a backend server does not respond and takes it
temporarily out of the list of active servers. If you've got enough
redundancy nobody will notice any difference.
> On 5/9/06, Harmen <ha...@tty.nl> wrote:
> >On Tue, May 09, 2006 at 04:47:14PM +0800, Ken Perl wrote:
> >> Is it possible to make a modperl application to run in cluster? if
> >> yes, how to do that? any doc?
> >
> >Take a few mod_perl servers and put one pound (http://www.apsis.ch/poung/)
> >reverse
> >proxy in front of them.
> >Works great.
--
The Moon is Waxing Gibbous (86% of Full)
Re: clusters
Posted by Ken Perl <ke...@gmail.com>.
Do you mean that the pound work great with apache 2.0 and apache 1.3 both?
On 5/9/06, Harmen <ha...@tty.nl> wrote:
> On Tue, May 09, 2006 at 04:47:14PM +0800, Ken Perl wrote:
> > Is it possible to make a modperl application to run in cluster? if
> > yes, how to do that? any doc?
>
> Take a few mod_perl servers and put one pound (http://www.apsis.ch/poung/) reverse
> proxy in front of them.
> Works great.
>
> --
> The Moon is Waxing Gibbous (85% of Full)
>
--
perl -e 'print unpack(u,"62V5N\"FME;G\!E<FQ`9VUA:6PN8V]M\"\@``
")'
Re: clusters
Posted by Harmen <ha...@tty.nl>.
On Tue, May 09, 2006 at 04:47:14PM +0800, Ken Perl wrote:
> Is it possible to make a modperl application to run in cluster? if
> yes, how to do that? any doc?
Take a few mod_perl servers and put one pound (http://www.apsis.ch/poung/) reverse
proxy in front of them.
Works great.
--
The Moon is Waxing Gibbous (85% of Full)
Re: clusters
Posted by "E.Q. McGoon" <ok...@baimac.com>.
Jonathan Vanasco wrote:
> Chiming in on everything thats been said:
>
> Proxy
> You can also use lighttpd for proxying. I do and find it works well
> http://lighttpd.org
>
> Cluster DB
> Postgresql has slony for replication
> http://gborg.postgresql.org/project/slony1/projdisplay.php
>
> Session / Scaling / Caching
<snip>
This is good stuff <g> Thanks to you and Perrin.
--
Re: clusters
Posted by Jonathan Vanasco <mo...@2xlp.com>.
Chiming in on everything thats been said:
Proxy
You can also use lighttpd for proxying. I do and find it works well
http://lighttpd.org
Cluster DB
Postgresql has slony for replication
http://gborg.postgresql.org/project/slony1/projdisplay.php
Session / Scaling / Caching
I've found this works well:
code from the start with a min of 3 db handles - read , write ,
session. should your app be small, let them all have the same
connect args, but make sure your internal logic is addressing them by
db. you may want more handles to dole permissions on the read/write
- fine. but don't ever start an app with less than 3 handles -
you'll want to kill yourself when you try to scale.
put sessions in a separate db. if possible, put them on a separate
machine. i only use postgres now for application data, but I still
use mysql for sessions on a sep. machine -- its super fast as isam or
innodb for stuff like that.
i use memcached extensively. like others here said, you can't
realistically cache page content -- but you can cache the data used
to assemble it. in one app I have a lot of super complex joins -
12-20 tables on average. i optimized my application overall by
pulling in more data off the join than I need so its a bit more
generalized, then storing the resultset as a perl hash in memcached
with a 10min ttl. if anything that would affect the join is
inserted, i just send a clear to memcached for that key. pages are
built against the hash if it exists, or a live db call + cache set if
it doesn't. it puts more work on perl - but a completely negligible
amount -- maybe 3% more execution time and memory on the page
generation. But I'm saving a ton of db resources as I don't directly
query as much - so my db time is cut sharply. i'd say this results
in about a overall 50% speedup on page generation because of that.
| - - - - - - - - - - - - - - - - - - - -
| RoadSound.com / Indie-Rock.net
| Collaborative Online Management And Syndication Tools
| - - - - - - - - - - - - - - - - - - - -
Re: clusters
Posted by Perrin Harkins <pe...@elem.com>.
On Tue, 2006-05-09 at 17:51 +0000, E.Q. McGoon wrote:
> Now I'm starting to see where my mental blocks are. I've always
> operated under the notion that sessions are best left to the operating
> system on each node, so to speak, in an effort to reduce db traffic.
Once you have a cluster, sessions need to be shared between machines.
People often try to invent ways of doing this that will not use a
database, like putting them on NFS. Then they have two problems:
scaling the database and scaling NFS.
Using the same storage technology for your sessions that you use for all
your other shared data means that you have only one thing that you have
to make reliable and learn to scale. Since most sites already need the
database to be up all the time, it's an obvious choice.
- Perrin
Re: clusters
Posted by "E.Q. McGoon" <ok...@baimac.com>.
Perrin Harkins wrote:
> On Tue, 2006-05-09 at 17:12 +0000, E.Q. McGoon wrote:
> > Certainly, but I'm struggling with the "big picture" so to speak.
> > How to plan a session handling strategy that will grow with the
> > application as more hardware is added to accommodate growing usage.
>
> Why is this any different from your general database strategy? If the
> database can't handle the load, having sessions still working is
> useless.
> Strategies for scaling databases are widely discussed. The basic
> progression is from one big database to one write database and several
> replicated read databases and then to partitioned data (e.g. users A-H
> on this database, I-N on another, etc.).
Now I'm starting to see where my mental blocks are. I've always
operated under the notion that sessions are best left to the operating
system on each node, so to speak, in an effort to reduce db traffic.
> You might be interested in this series of blog posts by Tim O'Reilly
> on the subject:
> http://radar.oreilly.com/archives/2006/05/brian_aker_of_mysql_responds
> .html
I will read this. Thanks for your feedback.
--
Re: clusters
Posted by Perrin Harkins <pe...@elem.com>.
On Tue, 2006-05-09 at 17:12 +0000, E.Q. McGoon wrote:
> Certainly, but I'm struggling with the "big picture" so to speak. How
> to plan a session handling strategy that will grow with the application
> as more hardware is added to accommodate growing usage.
Why is this any different from your general database strategy? If the
database can't handle the load, having sessions still working is
useless.
Strategies for scaling databases are widely discussed. The basic
progression is from one big database to one write database and several
replicated read databases and then to partitioned data (e.g. users A-H
on this database, I-N on another, etc.).
You might be interested in this series of blog posts by Tim O'Reilly on
the subject:
http://radar.oreilly.com/archives/2006/05/brian_aker_of_mysql_responds.html
- Perrin
Re: clusters
Posted by "E.Q. McGoon" <ok...@baimac.com>.
JupiterHost.Net wrote:
>
>
> E.Q. McGoon wrote:
> > Perrin Harkins wrote:
> >
> > <snip>
> >
> > > Tons of them. Practically any doc about clustering will apply. I
> > > wrote one ages ago about one mod_perl site I worked at:
> > > http://perl.apache.org/docs/tutorials/apps/scale_etoys/etoys.html
> > >
> > > Is there some specific part you're having trouble getting your
> > > head around?
> >
> >
> > Can you elaborate further on your session management strategy in the
>
> have you looked at CGI::Session or Apache::Session ??
Certainly, but I'm struggling with the "big picture" so to speak. How
to plan a session handling strategy that will grow with the application
as more hardware is added to accommodate growing usage.
--
Re: clusters
Posted by "JupiterHost.Net" <ml...@jupiterhost.net>.
E.Q. McGoon wrote:
> Perrin Harkins wrote:
>
> <snip>
>
>>Tons of them. Practically any doc about clustering will apply. I
>>wrote one ages ago about one mod_perl site I worked at:
>>http://perl.apache.org/docs/tutorials/apps/scale_etoys/etoys.html
>>
>>Is there some specific part you're having trouble getting your head
>>around?
>
>
> Can you elaborate further on your session management strategy in the
have you looked at CGI::Session or Apache::Session ??
Re: clusters
Posted by Perrin Harkins <pe...@elem.com>.
On Tue, 2006-05-09 at 14:00 +0000, E.Q. McGoon wrote:
> Can you elaborate further on your session management strategy in the
> article?
We used a lightly hacked Apache::Session, storing data in Oracle. We
used mod_unique_id to generate IDs. We sent the IDs in cookies, along
with a simple authentication hash using Digest::SHA1. I recommend not
putting very much in your sessions, e.g. put the ID of the shopping cart
in, not the whole shopping cart, and store the data in database tables.
Now that CGI::Session has been taken over by new maintainers, it's been
improving a lot. You might want to look there first these days.
> Is there anything you would do differently now with the availability of MP2?
I would use an output filter for doing includes instead of hacking
Apache::SSI. I added a #template call that would run a TT template to
generate some dynamic content. That would be easier to do now with a
filter.
> Our team is currently contemplating a project involving a high volume
> of web services and dynamic page views. We have a large Perl code base
> that supports the mission of the application and would like to have a
> large MP2 app under our belt instead of reverting back to PHP/Zend.
Many large sites run on Perl, so I don't see any reason you would need
to switch. If there's something you're having trouble with, go ahead
and ask.
- Perrin
Re: clusters
Posted by "E.Q. McGoon" <ok...@baimac.com>.
Perrin Harkins wrote:
<snip>
> Tons of them. Practically any doc about clustering will apply. I
> wrote one ages ago about one mod_perl site I worked at:
> http://perl.apache.org/docs/tutorials/apps/scale_etoys/etoys.html
>
> Is there some specific part you're having trouble getting your head
> around?
Can you elaborate further on your session management strategy in the
article? Is there any "step-by-step" information available? Is there
anything you would do differently now with the availability of MP2?
Our team is currently contemplating a project involving a high volume
of web services and dynamic page views. We have a large Perl code base
that supports the mission of the application and would like to have a
large MP2 app under our belt instead of reverting back to PHP/Zend.
--
Re: clusters
Posted by Harmen <ha...@tty.nl>.
On Tue, May 09, 2006 at 01:45:57PM -0400, Kevin A. McGrail wrote:
> Taking this another step further, I don't get the impression that perlbal or
> pound will achieve the "temporary" caching on the middle-proxy server
> allowing the back-end server to get back to other requests.
>
> Anyone know for sure?
Pound does that and it good at it. Not sure if it works for uploads as
well.
--
The Moon is Waxing Gibbous (88% of Full)
Re: clusters
Posted by Frank Wiles <fr...@wiles.org>.
On Tue, 9 May 2006 13:45:57 -0400
"Kevin A. McGrail" <km...@pccc.com> wrote:
> Taking this another step further, I don't get the impression that
> perlbal or pound will achieve the "temporary" caching on the
> middle-proxy server allowing the back-end server to get back to other
> requests.
>
> Anyone know for sure?
I know perlbal does "cache" the result from the backend server and
spoon feeds it out to the client.
---------------------------------
Frank Wiles <fr...@wiles.org>
http://www.wiles.org
---------------------------------
Re: clusters
Posted by "Kevin A. McGrail" <km...@pccc.com>.
Taking this another step further, I don't get the impression that perlbal or
pound will achieve the "temporary" caching on the middle-proxy server
allowing the back-end server to get back to other requests.
Anyone know for sure?
Sincerely,
KAM
> On Tue, 2006-05-09 at 13:24 -0400, Kevin A. McGrail wrote:
> > My understand is that some people do caching on B simply to let A spit
out
> > all the data and let B handle the connection for as long as it takes to
> > perform the transfer.
>
> That's right. It's very useful for images and multimedia files too.
Re: clusters
Posted by Perrin Harkins <pe...@elem.com>.
On Tue, 2006-05-09 at 13:24 -0400, Kevin A. McGrail wrote:
> My understand is that some people do caching on B simply to let A spit out
> all the data and let B handle the connection for as long as it takes to
> perform the transfer.
That's right. It's very useful for images and multimedia files too.
> I've heard that this can have excellent benefits but have no real world
> examples or experience handy.
I've seen it cut down the number of database connections needed by a
factor of 10. It really helps.
- Perrin
Re: clusters
Posted by "Kevin A. McGrail" <km...@pccc.com>.
> As a caveat to the E-mails about caching pages using a proxy, the
usefulness
> of such depends on how much traffic your site will see to static pages.
In
My understanding doesn't agree with this. The point about caching also
deals with dynamic one time through away pages in a scenario as described
below:
Server A -> Proxy Server B -> Users 1 through 50
User 1 requests data. Server B interecepts and sends to Server A, gets
answer and returns to 1.
If B cannot cache the data, does it keep a connection open to A as long as 1
takes to receive the data?
For example, if 50 14.4Kbps dialup users come and take 10 minutes to
download a file, is Server A & Server B tied up with 50 connections for 10
minutes?
My understand is that some people do caching on B simply to let A spit out
all the data and let B handle the connection for as long as it takes to
perform the transfer.
I've heard that this can have excellent benefits but have no real world
examples or experience handy.
Regards,
KAM
Re: clusters
Posted by Len Kranendonk <le...@primaat.com>.
> And also this:
> "NDB is an in-memory storage engine"
No longer true: As of MySQL 5.1.6 it's possible to store
tables on disk rather than in memory.
See http://dev.mysql.com/doc/refman/5.1/en/mysql-cluster-disk-data.html
But, I wouldn't recommend using MySQL cluster to power your
high traffic website. I've tested it but the throughput is at least 10 -
100x
worse in compare to a good old MySQL master / slave setup on the
same hardware. Esspecially the writes are slow.
SQL Relay offers load balancing functionality, without having to change
your existing code: http://sqlrelay.sourceforge.net/
Len
Re: clusters
Posted by Harmen <ha...@tty.nl>.
On Tue, May 09, 2006 at 12:51:17PM -0400, Perrin Harkins wrote:
> On Tue, 2006-05-09 at 18:41 +0200, Harmen wrote:
> > It _is_ stored on disk, save for powerfailures, but all the data needs to
> > fit in ram. So it's limited to the amount of ram you can give mysql.
>
> It is on disk? This quote from the manual led me to think it wasn't:
> "MySQL Cluster does not support durable commits on disk. Commits are
> replicated, but there is no guarantee that logs are flushed to disk on
> commit."
Hmmm, from the faq it seems they do 'some effort' to log the transactions. But there
realy are no guarantees. Not really usefull, yet, indeed.
>
> And also this:
> "NDB is an in-memory storage engine"
>
> - Perrin
>
--
The Moon is Waxing Gibbous (88% of Full)
Re: clusters
Posted by Perrin Harkins <pe...@elem.com>.
On Tue, 2006-05-09 at 18:41 +0200, Harmen wrote:
> It _is_ stored on disk, save for powerfailures, but all the data needs to
> fit in ram. So it's limited to the amount of ram you can give mysql.
It is on disk? This quote from the manual led me to think it wasn't:
"MySQL Cluster does not support durable commits on disk. Commits are
replicated, but there is no guarantee that logs are flushed to disk on
commit."
And also this:
"NDB is an in-memory storage engine"
- Perrin
Re: clusters
Posted by Harmen <ha...@tty.nl>.
On Tue, May 09, 2006 at 12:32:41PM -0400, Perrin Harkins wrote:
> On Tue, 2006-05-09 at 11:28 -0500, Jeremy Brooks wrote:
> > There's also clustering in MySQL 5 that looks very enticing to me as an
> > additional layer or scale out and redundancy. It's not your typical
> > master/slave scenario and actaully allows for real load balancing and
> > failover.
>
> It's a pretty limited system, as described here:
> http://dev.mysql.com/doc/refman/5.0/en/mysql-cluster-limitations.html
>
> No referential integrity and no disk storage (!) are the main issues for
It _is_ stored on disk, save for powerfailures, but all the data needs to
fit in ram. So it's limited to the amount of ram you can give mysql.
> me. It might be okay for transient data that you don't care about
> losing if the power goes out.
>
> - Perrin
>
--
The Moon is Waxing Gibbous (88% of Full)
RE: clusters
Posted by Jeremy Brooks <jb...@austin.rr.com>.
It certainly is less appealing after reading that. I've only read a limited
amount about how it functions. Thank you for pointing that out Perrin.
-Jeremy
> -----Original Message-----
> From: Perrin Harkins [mailto:perrin@elem.com]
> Sent: Tuesday, May 09, 2006 11:33 AM
> To: Jeremy Brooks
> Cc: 'Ken Perl'; modperl@perl.apache.org
> Subject: RE: clusters
>
> On Tue, 2006-05-09 at 11:28 -0500, Jeremy Brooks wrote:
> > There's also clustering in MySQL 5 that looks very enticing
> to me as
> > an additional layer or scale out and redundancy. It's not your
> > typical master/slave scenario and actaully allows for real load
> > balancing and failover.
>
> It's a pretty limited system, as described here:
> http://dev.mysql.com/doc/refman/5.0/en/mysql-cluster-limitations.html
>
> No referential integrity and no disk storage (!) are the main
> issues for me. It might be okay for transient data that you
> don't care about losing if the power goes out.
>
> - Perrin
RE: clusters
Posted by Perrin Harkins <pe...@elem.com>.
On Tue, 2006-05-09 at 11:28 -0500, Jeremy Brooks wrote:
> There's also clustering in MySQL 5 that looks very enticing to me as an
> additional layer or scale out and redundancy. It's not your typical
> master/slave scenario and actaully allows for real load balancing and
> failover.
It's a pretty limited system, as described here:
http://dev.mysql.com/doc/refman/5.0/en/mysql-cluster-limitations.html
No referential integrity and no disk storage (!) are the main issues for
me. It might be okay for transient data that you don't care about
losing if the power goes out.
- Perrin
RE: clusters
Posted by Jeremy Brooks <jb...@austin.rr.com>.
There's also clustering in MySQL 5 that looks very enticing to me as an
additional layer or scale out and redundancy. It's not your typical
master/slave scenario and actaully allows for real load balancing and
failover.
> -----Original Message-----
> From: Jeremy Brooks [mailto:jbrooks4@austin.rr.com]
> Sent: Tuesday, May 09, 2006 11:19 AM
> To: 'Jeremy Brooks'; 'Perrin Harkins'; 'Ken Perl'
> Cc: modperl@perl.apache.org
> Subject: RE: clusters
>
> As a caveat to the E-mails about caching pages using a proxy,
> the usefulness of such depends on how much traffic your site
> will see to static pages. In our situation, the only static
> pages are our "contact us" page, "privacy policy" ,etc.
> Those pages don't get enough traffic to justify the
> maintenance of setup overhead of a proxy. Our main page is
> not static and neither are the product and category pages.
> Actually over 99% of our pages are dynamic with rotating
> specials and such. The usefulness of caching on pages via a
> proxy also depends on your typical landing pages.
> If you have a customer click to your site to a page with
> dynamic content (specials, session IDs in form data, etc)
> then caching to the most frequently viewed pages is useless
> and even damaging/problematic.
>
> Regarding Round Robin DNS, this scenario will become a
> nightmare should you loose a server that is not redundant on
> the same IP. Even with a very short TTL, your customers will
> see more downtime then they like and may go elsewhere should
> you loose a server. DNS load balancing is generally not a
> wise load balancing method because it's NOT actually
> balancing load in real time.
> The affect of round robin DNS is something akin to breaking
> up load sometime last week and hoping it all works out today.
> It's not really an active solution since servers die and
> nothing changes in the customers view for quite sometime. In
> fact, TTL must be reached before other DNS servers update
> their cache and most importantly, your customers PC must
> actually begin to us the new IP to connect! So, now you're
> dependent on DNS client
> caching on your customers PCs... Not good. On a Windows PC
> this can mean
> restarting the browser, on an OSX machine this may mean
> restarting the damn machine (really, I've experienced this).
>
> Go with IPVS as a base and proxy if it makes sense. Use DNS
> load balancing only if you have redundancy on each node
> connected to each IP and if load balancing public IPs is to
> much pain otherwise.
>
> You can use IPVS in several different topologies or you can
> mix them to get what you want.
>
> http://www.ultramonkey.org/3/topologies/
>
>
> -Jeremy Brooks
>
>
> > -----Original Message-----
> > From: Jeremy Brooks [mailto:jbrooks4@austin.rr.com]
> > Sent: Tuesday, May 09, 2006 10:26 AM
> > To: 'Perrin Harkins'; 'Ken Perl'
> > Cc: modperl@perl.apache.org
> > Subject: RE: clusters
> >
> > We use Ultra Monkey (IPVS) to cluster our E-com site with commodity
> > hardware.
> > Your application needs to cache sessions in a common
> database table in
> > order to make session work consistantly.
> >
> >
> >
> > > -----Original Message-----
> > > From: Perrin Harkins [mailto:perrin@elem.com]
> > > Sent: Tuesday, May 09, 2006 7:28 AM
> > > To: Ken Perl
> > > Cc: modperl@perl.apache.org
> > > Subject: Re: clusters
> > >
> > > Ken Perl wrote:
> > > > Is it possible to make a modperl application to run in cluster?
> > >
> > > Of course.
> > >
> > > > if yes, how to do that?
> > >
> > > The same as any other web system -- multiple machines, a load
> > > balancer, shared data on a database or other system.
> > >
> > > > any doc?
> > >
> > > Tons of them. Practically any doc about clustering will
> apply. I
> > > wrote one ages ago about one mod_perl site I worked at:
> > > http://perl.apache.org/docs/tutorials/apps/scale_etoys/etoys.html
> > >
> > > Is there some specific part you're having trouble getting
> your head
> > > around?
> > >
> > > - Perrin
>
RE: clusters
Posted by Jeremy Brooks <jb...@austin.rr.com>.
As a caveat to the E-mails about caching pages using a proxy, the usefulness
of such depends on how much traffic your site will see to static pages. In
our situation, the only static pages are our "contact us" page, "privacy
policy" ,etc. Those pages don't get enough traffic to justify the
maintenance of setup overhead of a proxy. Our main page is not static and
neither are the product and category pages. Actually over 99% of our pages
are dynamic with rotating specials and such. The usefulness of caching on
pages via a proxy also depends on your typical landing pages.
If you have a customer click to your site to a page with dynamic content
(specials, session IDs in form data, etc) then caching to the most
frequently viewed pages is useless and even damaging/problematic.
Regarding Round Robin DNS, this scenario will become a nightmare should you
loose a server that is not redundant on the same IP. Even with a very short
TTL, your customers will see more downtime then they like and may go
elsewhere should you loose a server. DNS load balancing is generally not a
wise load balancing method because it's NOT actually balancing load in real
time.
The affect of round robin DNS is something akin to breaking up load sometime
last week and hoping it all works out today. It's not really an active
solution since servers die and nothing changes in the customers view for
quite sometime. In fact, TTL must be reached before other DNS servers
update their cache and most importantly, your customers PC must actually
begin to us the new IP to connect! So, now you're dependent on DNS client
caching on your customers PCs... Not good. On a Windows PC this can mean
restarting the browser, on an OSX machine this may mean restarting the damn
machine (really, I've experienced this).
Go with IPVS as a base and proxy if it makes sense. Use DNS load balancing
only if you have redundancy on each node connected to each IP and if load
balancing public IPs is to much pain otherwise.
You can use IPVS in several different topologies or you can mix them to get
what you want.
http://www.ultramonkey.org/3/topologies/
-Jeremy Brooks
> -----Original Message-----
> From: Jeremy Brooks [mailto:jbrooks4@austin.rr.com]
> Sent: Tuesday, May 09, 2006 10:26 AM
> To: 'Perrin Harkins'; 'Ken Perl'
> Cc: modperl@perl.apache.org
> Subject: RE: clusters
>
> We use Ultra Monkey (IPVS) to cluster our E-com site with
> commodity hardware.
> Your application needs to cache sessions in a common database
> table in order to make session work consistantly.
>
>
>
> > -----Original Message-----
> > From: Perrin Harkins [mailto:perrin@elem.com]
> > Sent: Tuesday, May 09, 2006 7:28 AM
> > To: Ken Perl
> > Cc: modperl@perl.apache.org
> > Subject: Re: clusters
> >
> > Ken Perl wrote:
> > > Is it possible to make a modperl application to run in cluster?
> >
> > Of course.
> >
> > > if yes, how to do that?
> >
> > The same as any other web system -- multiple machines, a load
> > balancer, shared data on a database or other system.
> >
> > > any doc?
> >
> > Tons of them. Practically any doc about clustering will apply. I
> > wrote one ages ago about one mod_perl site I worked at:
> > http://perl.apache.org/docs/tutorials/apps/scale_etoys/etoys.html
> >
> > Is there some specific part you're having trouble getting your head
> > around?
> >
> > - Perrin
RE: clusters
Posted by Jeremy Brooks <jb...@austin.rr.com>.
We use Ultra Monkey (IPVS) to cluster our E-com site with commodity
hardware.
Your application needs to cache sessions in a common database table in order
to make session work consistantly.
> -----Original Message-----
> From: Perrin Harkins [mailto:perrin@elem.com]
> Sent: Tuesday, May 09, 2006 7:28 AM
> To: Ken Perl
> Cc: modperl@perl.apache.org
> Subject: Re: clusters
>
> Ken Perl wrote:
> > Is it possible to make a modperl application to run in cluster?
>
> Of course.
>
> > if yes, how to do that?
>
> The same as any other web system -- multiple machines, a load
> balancer, shared data on a database or other system.
>
> > any doc?
>
> Tons of them. Practically any doc about clustering will
> apply. I wrote one ages ago about one mod_perl site I worked at:
> http://perl.apache.org/docs/tutorials/apps/scale_etoys/etoys.html
>
> Is there some specific part you're having trouble getting
> your head around?
>
> - Perrin
Re: clusters
Posted by Perrin Harkins <pe...@elem.com>.
Ken Perl wrote:
> Is it possible to make a modperl application to run in cluster?
Of course.
> if yes, how to do that?
The same as any other web system -- multiple machines, a load balancer,
shared data on a database or other system.
> any doc?
Tons of them. Practically any doc about clustering will apply. I wrote
one ages ago about one mod_perl site I worked at:
http://perl.apache.org/docs/tutorials/apps/scale_etoys/etoys.html
Is there some specific part you're having trouble getting your head around?
- Perrin