ServeTheHome 2013 Architecture Discussion

[Patrick]

Yea if you want open stack, stackops is the best choice, but its still a PITA. And I dont think you really get much advantages. NAS/SAN integration is difficult at best.

Eucalyptus - Havent Tried
CloudStack - Relatively Easy to Install - I got lost in the networking though. Requires either Citrix, KVM, or ESX w/ vCenter
OnApp - Love the interface and iphone apps, really cool product, but it is expensive. Min $500/mo 6k/yr.
ProxMox - Havent Tried

So on the CloudStack thing, looks easy, but 12.04 LTS installation is still broken. Not sure how I like a Ubuntu 10.10 release underpinning.

ProxMox I managed to get two boxes up in ~15 minutes.

Really amazing to me. Should be insert installation media over IPMI on all of the nodes. Boot to installation media. Run one installer on the "master". Distributed file system across nodes installed. Backup storage drives marked. Installer sets up basic networking, firewall in front of the cloud, HA managers and etc. Want to add a box? Run the boot CD, select "Add" as an installation option and have everything auto configure. Somewhat amazing nobody has done this yet.

 

[nitrobass24]

Yea i agree, most of these are way too difficult for basic installations.

Re Cloudstack: I didnt know Ubuntu was broken, but personally I am not a fan of Debian distros. I prefer RHEL/CentOS. Not sure its necessarily better, but the companies I do work for use RHEL so that is what I am most familiar with.

On a side note I am amazed every day how many people are still running HPUX and AIX. Mainstream Linux is so much better!

ProxMox - Is that KVM only? Can you run w/o support subscription? I am finding their website to be a little cryptic regarding their offering.

 

[Patrick]

Proxmox is community support + paid commercial support. KVM and OpenVZ.

On the Ubuntu thing, STH uses an Amazon AMI (RHEL based). I use Ubuntu for most other stuff though.

I swear... someone is going to figure this out and make a fortune (maybe its OnApp with storage.)

 

[Jeggs101]

Did you decide Fiberhub or Dacentec? Want to see which 1 you went with + why.

Hope you are writing all of this up.

 

[Patrick]

More info soon. Flight booked to the DC Saturday.

 

[Patrick]

Colocation provider inked. Main site post on reasoning.

 

[PigLover]

Patrick - have a great time in Vegas. Hope the integration goes so smoothly that you have plenty of time to donate money at the Casino's!

 

[Patrick]

Patrick - have a great time in Vegas. Hope the integration goes so smoothly that you have plenty of time to donate money at the Casino's!

I think the letters BGP will ensure that I don't have extra time!

Thinking that the forums may get migrated first, then the main site. Not 100% on that one, but current assumption is that this might work.

 

[Patrick]

Posting from laptop connected through the colocation switch and the first pfsense box

Hello web!

 

[nitrobass24]

Nice I like it.

 

[MiniKnight]

Doesn't seem to be any faster. Is this all done now?

 

[Patrick]

Not done at all.

Status update:
Proxmox cluster is live.
carp on the two pfsense nodes is being setup

Rack911 is working hard on the site. Will post in the main thread.

 

[nitrobass24]

I guess I missed it, but what are you using as boot devices for the nodes?

I am building out a citrix setup at home, and am trying to figure out what I want to use to boot the hypervisors.
Want something small, cheap. Maybe like a SATA DOM or SD card.

 

[Patrick]

Using various SSDs. Real reason is that I had many very lightly used ones.

 

[cafcwest]

I saw backups mentioned early in the discussion, but didn't see anything about since (of course I could be missing it).

How are you backing up your nodes?

 

[Jeggs101]

I saw backups mentioned early in the discussion, but didn't see anything about since (of course I could be missing it).

How are you backing up your nodes?

Likely the 3TB disks.

 

[RimBlock]

Loved the colo article.

Couple of questions.
I see you now have all your prod nodes in a single chassis. Why this rather than having one in each of the two live chassis ?.
Any reason for not connecting up the 3rd unit ?. I appreciate it will use some power but you would be able to access it as and when you required should a fault develop with any of the other chassis/nodes. You could even do a daily refresh for HA and then power off.

RB

 

[Patrick]

Two reasons. First the idea is it is a cold spare . Since the only thing it needs is a quick 5 min swap. Second from a power perspective it is all sitting at 5A. Trying to keep it there since it costs more to go over.

 

[RimBlock]

Two reasons. First the idea is it is a cold spare . Since the only thing it needs is a quick 5 min swap. Second from a power perspective it is all sitting at 5A. Trying to keep it there since it costs more to go over.

Yep the power is of concern if you are right on the edge.

Who will be wiring up the cold spare should it be needed ?. I was under the impression that you are usually in another state.

Does connecting a single rather than both redundant PSUs on one of these servers make a difference to the power figures ?. If so then maybe having the core prod server on both redundant PSUs and the test / spare on a single one. Just a thought . I personally hate the idea of being in a position of having to wait for others if there is a prod issue so like to try and make sure I have all the access to the things I need without having to involve others wherever possible.

I hope to have a couple coming over to me soon and the upgrades you are doing for the site and forums are all very interesting so thanks for taking the time to put them up.

RB

 

[Patrick]

The idea is that each chassis is redundant if one fails. Will likely need remote hands but really easy to do given that everything is hot swap.