├── .gitignore └── 2014-02-06 ├── poster-surf.pdf └── topics.md /.gitignore: -------------------------------------------------------------------------------- 1 | $.* 2 | -------------------------------------------------------------------------------- /2014-02-06/poster-surf.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/sagemath/office-hours/master/2014-02-06/poster-surf.pdf -------------------------------------------------------------------------------- /2014-02-06/topics.md: -------------------------------------------------------------------------------- 1 | # SageMathCloud virtual office hours -- Feb 6, 2014 2 | 3 | Video: 4 | 5 | ## Status 6 | 7 | ### Backend -- stuff you hopefully won't notice much! 8 | 9 | - **ZFS backend architecture**: moved from ext4 on individual VM's + snapshots made using bup(=Python+git) to... ZFSonlinux with two replicas per data center, distributed using "consistent hashing". 10 | - took most of December; some of January. 11 | - spent a lot of time investigating and building another(geo-)distributed file system options on glusterfs, which I threw away 12 | - some growing pains with ZFS, but is working pretty well 13 | - finally we have high availability: if up to 5 machines that host your project fail, you can continue working 14 | - snapshots are local, so in the long run much MORE can be done with them -- much, much faster to browse. 15 | - open question: save all snapshots or trim old ones: e.g., save only hourly, daily, weekly 16 | - More systematic backend monitoring 17 | - Major VPN/network improvements (tinc scalability) 18 | - Google Compute Engine 19 | 20 | ### Usage 21 | 22 | - Increasing: number of accounts doubled in 89 days: today "There are 20310 accounts and 30183 projects." 23 | - Yesterday was maybe heaviest ever: 350-400 simultaneous clients most of the day. 24 | - If things feel slow, and your net connection is good, the most likely cause is somebody using a lot of resources on the same VM as you, e.g., a fork bomb, bitcoin mining, etc. 25 | - Very near future: cgroups, to allocate resources more fairly. 26 | 27 | ### Marketing 28 | 29 | - JMM in Baltimore -- booth 30 | - Poster -- see poster-surf.pdf that Hao Chen made. 31 | - Harald has been tracking websites that point to SMC -- lots of interesting courses. 32 | 33 | ### Commercialization 34 | 35 | - not yet ready, due to more needing to be implemented (e.g., cgroups) for this to make sense 36 | - not clear what/when/how to sell 37 | - first product will be a "private premium plan" available only to a select group of users we'll choose (and sold by Univ of Washington). 38 | - I will focus much more on commercialization starting in June 2014 (when my sabbatical starts) 39 | --------------------------------------------------------------------------------