On Thu, Sep 2, 2010 at 3:42 PM, Paul Cutler <[email protected]> wrote: > Hi, > > I wanted to follow-up quickly on a discussion we were just having IRC. > Germán recently sent an email to foundation-list asking teams for > feedback if they need any funds budgeted for next fiscal year which > starts in October. > > Thanks to Jeff's donation of Combobox earlier this year we have a shiny > new server we need to utilize more. With the issues with Label today, I > would like to ask a couple questions: > > * What would we like to migrate to Combobox?
Well originally, I wanted to buy a new database server. Drawable appeared and killed that idea. After talking to owen it was decided that container would be a good box to retire. So... I bought combobox with the idea of moving some vms to it and having it be the primary storage for all of the gnome infrastructure in the PHX redhat facility. > * What server hardware (and configuration) would be optimal for the > Foundation, considering most of our hardware is now out of support > contracts? (How many new servers hosting what services?) 1.) Ideally, another server with stats comparable to combobox for more vms and builders. For instance, fpeters's report generation scripts kill 1 of the boxes (window or fixed can't remember which) regularly and eat all of the memory. That box runs a few other services if I recall. I would be great to give him a few decent vms to beat up all day for those tasks. Having a few dedicated build boxes is always a good thing as well. Additionallyu 2.) 1 lower end server that is still fast to run less resource hungry but still important services. To split up failure domains, it is wise to put nfs + ldap/dns on separate servers so that if 1 server tanks, you don't lose everything. Due to the chicken / egg problem, ldap, dns, and nfs should be on real hardware and not on virtual machines. I originally bought combobox for nfs. It would be nice to have another box for vms and misc projects that need horsepower and then a lower end but still decent new box for ldap/dns and things like perhaps mango and civicrm. As you know, some projects such as blip (http://blip.paulcutler.org) are great and might eventually have gnome hosted versions. Additionally, snowy (for tomboy online), and the inevitable getting things gnome web based version will be gnome hosted at some point. We need the infrastructure to provide these services for our users. The current setup is unsuitable for doing this well. > * What would our backup plan be if we could only buy one server? Well only buying 1 server is better than 0 but 1 really nice one and 1 lower end one would be perfect. The benefit outweights the downsides in my professional opinion here. The "backup plan" at a bare minimum would be to split ldap/dns to the backup server from label and menubar respectively. This is my opinion only and is up to the team ultimately. This would allow us to potentially retire 2 out of support servers with a newer faster and more energy efficient one. We could figure out the details should that be the case. > * What are the risks with the current setup? Some of the drives in critical gnome servers are a bit flaky and the hardware is out of support. If it breaks, we're screwed! Things like git or any of the web services such as wiki could go down and be down for long periods of time. > That's off the top of my head. Please discuss! :) > > Paul I thought (incorrectly) that in spending my own personal $$ to buy combobox, the foundation would front the money to buy one of equal or greater value. Perhaps I was a bit naive in thinking this without clarification, but I assumed this was the case. Ping me _offline_ for my reasoning on this if you've got questions why. We desperately need at least 1 new server to retire some of the older hardware such as label which is performing critical functions such as ldap master. If that goes down, things like git checkings will stop. We are going to be rolling out some changes fairly soon to make our ldap setup more robust, but for now it is what it is. TL;DNR We need at very least 1 new server if we want to ensure gnome services remain available at all times. Two would be ideal so we can separate out critical services. When label recently went offline and spent a looooong time coming up, we lost the following services: - jabber - live.gnome.org aka "the wiki" - piwik - which would cause all other gnome websites with piwik embedded to load more slowly - ldap - which breaks most everything. It causes fun things like preventing any git checkins - CiviCRM - which is critical for Stormy + Rosanna from my understanding - something I missed surely Thats just a good recent example. Combobox is a Dell 1U server which I'll have to recommend as they are better bang for the buck without skimping quality. The ILO stuff isn't as schmexy as the HP or Sun^WOracle servers, but the price point is right and the quality is very good. They extended warranties are also well cheap. [1] http://live.gnome.org/Sysadmin/Servers#combobox.gnome.org -- Jeff Schroeder Don't drink and derive, alcohol and analysis don't mix. http://www.digitalprognosis.com _______________________________________________ gnome-infrastructure mailing list [email protected] http://mail.gnome.org/mailman/listinfo/gnome-infrastructure
