New linuxha.net cluster over the next couple of weeks

[micoots]

Hi Simon,

I'm currently in the planning stages but very soon will start building my new linuxha.net clusters to move them into a high level data centre.

I've already decided that I will not be using my current clusters and moving them, instead I'll be buying two new servers, building linuxha.net on them and then only moving those.

My current clusters comprise of:

linuxha.net 1.0.11 for one apache cluster
linuxha.net 1.0.11 for one mysql cluster

As I'm only buying two machines, then I'm going to have both mysql and apache running on the same server. I'm buying two HP Proliant DL140's (1U) which only have two mirrored disks.

So although I'd love to have one node run apache, and the other node (standby node) run mysql - so each fails-over to the other one upon failure - with only 2 mirrored drives per machine, I doubt I can do this.

So I expect that both apache and mysql will run on one node and the other standby will sit there and do very little.

I'll likely be ordering the two servers this week.

Is there any recommendations you can make for me?

ie. should I go through an install of 1.0.11 as I'm familiar with it, or should I try building/installing 1.2.x ? I don't know whether that is production ready?

Note that the new data centre is major, one of the best in Australia and the link speed will be 100mb/100mb dedicated, so performance improvement through the system will be realised.

But as it's very expensive for me to visit the site when things go wrong, I'll need a rock solid environment which doesn't break down.

1.0.x has been very good for me over the years so I'm not sure if I should be leaving it.

Your advice is appreciated.

Michael.

[linuxha]

Hi Michael,
Firstly you can have an active:active configuration. I guess you would fdisk the two devices something like [as primary/extended partitions]

p /boot type linux
e / type raid autodetect
e swap type raid autodetect
e mysql type raid autodetect
e apache type raid autodetect

Then use mdadm to create 4 raid1 decides against the two disks [for /, and SWAP, and two empty md's [say md2 and md3] for example.

Simply create two volume groups using the free raid devices;

pvcreate /dev/md2; vgcreate apachevg /dev/md2
pvcreate /dev/md3; vgcreate mysqlcg /dev/md3

Do the same on each host and you can then use each Vg for each app - and run active:active if you so wish.

The point is that different volume groups can still reside on the same physical disks, and if apache has light I/O sharing the disks with a DB should not cause performance issues.

Other recommendations would be to stick with 1.0.x for the moment if you are happy with it. 1.2.x IS better, but I would suggest a testing phase first before such a production deployment!

Other recommendations are to have at least 2 net connections - on different subnets - one for DRBD traffic and one for public traffic. That will make a difference.

Regards,
Simon.