Server Planning Meeting

Attendees

• Brian M, Jared, Sheila

Goals & Summary

• To discuss the knowns/unknowns and issues around the current plan for deployment of Cosmo in order to support the 0.6 release goals for Chandler.
• The discussion touched on a number of areas around our policy for Cosmo and deployment strategy in general that this team was not prepared to answer.

Open Issues and owners (updated 20050811 by lisa)

• Do we need a separate dog food server internally at osaf to host the office calendar, conf room calendars etc and what is the plan for this?
  • Jared recommends a separate server.
  • Lisa would like to see OSAF dogfood in the same doghouse as the rest of the world -- the hosted service. Do we have privacy issues

• Role and relevance of FoxMark^? (Todd's bookmark sharing extension for FireFox^?) which could host a user's bookmark data on Cosmo
  • Jared doesn't foresee problems hosting bookmarks along with calendar data

• Who owns marketing communications here?
  • BCM will describe the "job" here to Pieter and see if he can take it on.

Decisions Made

• We need to have the out-of-the-box OSAF Sharing account, on Chandler installation, point to an instance of Cosmo rather than Pillikia around the M5 timeframe so we
• We should not be using Cosmo QA for this, we need to have some other instance of Cosmo.
• We will NOT install Chandler (continuous build or milestone) with a default server . We require the user to sign up for a user name and password, even if this is an extra step. Both Jared and Brian M would not like to continue the same setup as we did with Pillikia.
• Since we are not sure when we will have the server up and running and/or if M5 will be late, we may install Cosmo 0.2-02 (then upgrade to next release) or start by installing Cosmo 0.2-03. This is ok, we wil play it by ear.
• We will focus on getting one server up and running and then worry about whether or not we have 2 servers, multiple instances etc.
• We voted for no quota support for now.
• Jared thinks if we find someone abusing the server, we turn it off until we have the tools in place to control and monitor this.

More decisions Aug 11

• How many servers do we need?
  • We need a QA server managed by Aparna, a public hosted server hosted by Jared, and a development server to be the least stable of the three. Jared will make decisions about how to put these on same or different machines and which ports, etc. Let's worry just about the hosted server for now.

• What is Chandler 0.6 pointing to?
  • The host server that Jared runs

• Is this the same as subsequent milestones/builds/checkpoints in the 0.7 release? What are all the instances of Cosmo that we need?
  • We should just have one to start with and we can include a version number in the hostname in case later we have to have multiple servers for different versions of Chandler to point to. This might have to happen if we make changes in Cosmo that older versions of Chandler can't talk to for some reason.

• What are our policies for... ?
  • ie: data reliability - data preserved between updates?
  • support- what is expected.
  • accounts preserved between updates?
  • quotas?
  • what do we do for server abuse - ie:upload porn on the server
  • do we block certain mime types or extensions.
  • ANSWER: Generally Jared owns these policy decisions. Jared proposes 95% uptime; data may be wiped but rarely; we need to collect peoples email addresses in case of later use notifications...
  • Server abuse: ignorance is bliss -- we don't need to go hunting for porn or other abuses -- but any time abuse is pointed out to us we will simply wipe the account.
  • Is one possible strategy to turn off the server if we find someone abusing it? Yes.

• Are any policy decisions blocking? Space quotas are required for hosting so that we don't end up hosting mucho porn
  • We can't have space quotas -- even though they're fairly easy compared to bandwidth quotas -- by Aug 24
  • Developing space quotas will be a high prio for BCM after Aug 24

• What port does the server use?
  • Using port 80 on the hosted service makes it more usable for the consumers of the service -- it's the only port that can be assumed.
  • Using something other than port 80 as the default Cosmo port on installation makes that installation more usable for the person doing the installing -- it makes Cosmo more likely to work immediately (avoids conflicts with other software) and they can change it later if they're doing more than just testing Cosmo (and if they want to change it)
  • We're completely OK with using a different port on the hosted service than the Cosmo default install port.

• $$$ Who makes the budget decisions here and what are the considerations?
  • Jared proposes getting a new box with RAID 0, couple gigs of RAM, FC2 and Java -- existing boxes don't have enough RAM for hosted service. Mitch approved a round-numbers budget of around $1500.

Notes

• Chandler M5 currently scheuduled for Aug 24th (might be late) and M6 scheduled for Sept 14th (could be adjusted).
• Cosmo 0.2-03 on track for Aug 26th.
• Around the M5 time frame we would like to have the installation of Chandler point to Cosmo rather than Pillikia so we can start to test the sharing functionality and work out the issues in advance of the 0.6 release. Since many people within OSAF will be "dogfooding" the calendar, we need to have the latest sharing features accessible.
• We decided that using QA Cosmo for this is not a good option, we really need a demo cosmo server.
• Brian M and Jared also didn't want to configure the same setup as Pilikia with a pre-configured account setup with username and password. We can point to the right server out of the box but we need to have the user take the extra step and setup their own account.
• Sheila will own the communication for this once we switch over.

• Right now we don't really have a box available to use.
• Jared's plan is to purchase something, set it up and install it as ISC.
• He should be able to start that next week and has set a red flag for the 20th if this is not done.
• There is some dependency on budget here that Jared will check out. Not sure who ok's these decisions.
• Jared's expectations
  • 99.5% uptime
  • data will not get wiped out.
• Is there a distinction of which server is public and which is not. In a sense all are public since any build can be downloaded. Basically, 0.6 should always run. The milestones should always work and point to some version of Cosmo that can be updated. If this is the same server, fine but that's not Sheila's decision to make. The requirement is that 0.6 run but we have something to run the 0.7 release builds of Chandler against that will have upgrades of Cosmo.
• We will focus on getting one server up and running first and decide if this is the official public server or not.
• Need to decide
  • hostname -> cosmo-demo-02
  • port -> need someone to decide this
  • Cosmo version -> probably 0.2-02 but maybe 0.2-03 depending on timing.

• We talked a bit about the long term plan
  • Jared thought we could have a scalable plan in 3-4 months and define all the requirements.
  • Do we need to have an osaf calendaring server?