-   Linux - General (
-   -   HA/LB Clustering - best solution for Apache and Red5 ? (

rob840 06-16-2010 08:32 AM

HA/LB Clustering - best solution for Apache and Red5 ?
Firstly, I apologise if this not the correct forum for this subject but I do not see a "Linux - Clustering" forum, maybe there should be one?
In a similar vein, I am surprised not to see CentOs in "Linux - Distributions", is there a reason for this?

I am looking to build up a HA/LB linux cluster with specific software requirements;

For hardware, I have a number of dual xeon PE 2650s and would like to use them as efficiently as possible. These are 32bit systems, I anticipate scaling up to 64bit systems when I have a tested, working solution in place.

For distro, I am familiar with CentOs, Gentoo and Ubuntu but unsure as to which would be the best foundation, although leaning towards CentOs.

For software, I need to realise all the services provided by xampp (Apache, MySql, PHP, Perl, FTP), plus Red5 flash media server.

My current train of thought is;
6 physical servers; 2 Directors/Heartbeat, 2 Apache, 2 Red5
Gigabit private network for connecting the nodes.
CentOs 5.5 on all nodes.
DRDB across the 2 Apache nodes for Apache, MySql, PHP.
DRDB across the 2 Red5 nodes for Red5.

I welcome advice and comments from anyone with experience in this area to point me in the right direction, hopefully avoiding any gotchas or pitfalls that may be envisaged.
Many thanks in advance.

Blue_Ice 06-17-2010 07:31 AM

You probably also need OCFS2 or GFS (from Red Hat) on the systems with drbd installed. This is needed to be able to create primary/primary replication needed for load balancing. Normal filesystems can only be used with one system. So if you don't use OCFS2 or GFS, then you will use only one server. If you are going for CentOS, then I think you better use the GFS filesystem, because you can find it in the repositories.
Regarding MySQL, you might want to use the master/master replication within MySQL instead of replicating the database files with drbd.
I don't know much about Red5, so I am not able to tell much about it.
The 2 servers as used for the director and heartbeat is pretty standard. I don't know which director you are going to use though...

You can also look into the Red Hat Cluster Suite, I believe it is in CentOS as well.

I used the following links to get a problem like this solved, however it was more regarding making HA cluster for Xen:

PS: You might have wanted to post this in the Linux Server forum...

trey85stang 06-17-2010 06:00 PM

blue ice has hit the mail on the head, keep mysql off drbd and use active/active master/slave mysql replication. Everything you have said looks like a great starting point towards a working solution. Id stick with ocfs2, but thats just me.

Blue_Ice 06-18-2010 03:39 AM

I would stick with ocfs2 as well. But for RH like systems you might consider gfs. This is not without reason. If you upgrade the kernel you will have to manually install the right ocfs2 for your kernel as well. When you use gfs this is done at the same time with the kernel, at least when using yum...

rob840 06-23-2010 12:43 AM

Many thanks for the valued comments guys..

I have been doing some research on GFS and OCFS2 and have seen reports of some nasty issues with these,
such as; bugs in inode allocation when disk is very fragmented, racing conditions in file locking that leads to corruptions, bugs in distributed locking routines and fsck taking forever.

Due to these issues my preference is to go with GlusterFS.
However, this requires 64bit servers, so in the meantime I am going to try Heartbeat, DRDB and NFS for the file servers.

I had a look at CAOS NSA which appeared promising but gave up on it when the install process panicked.

Therefore, my new train of thought is;
4 physical servers;
2 Directors/Heartbeat each in xen with xampp stack and Red5 in dom0
2 Raid1 File servers with Heartbeat, DRDB and NFS exported to the dom0's
MySQL Multimaster Replication across the dom0's
Gigabit private network for connecting the nodes
CentOs 5.5 on all nodes

Hopefully this will achieve the same goal with less hardware than I originally anticipated.

I welcome any further advice and comments from anyone with experience in this area as I progress with this project.
Many thanks in advance.

Blue_Ice 06-23-2010 05:12 AM

Well, the main thing is for load balancing that the filesystem is capable of primary/primary replication. This way you don't need to setup scripts to synchronize your files.

All times are GMT -5. The time now is 09:40 PM.