Sg2009wc:cloud
From She's Geeky Wiki
Title: Cloud Computing
Session: 1-J Bay Area Jan 2009
Convener Sara Baker
Attendees:'
- Sarah Baker
- Stacy Banks
- Alison Chaiken
- Aminie Elsberry
- Jessica Nusbaum
- Nicole Poindexter
- Teresa Wood
- Gail Haspent
- Amy Bowers
- Lynn Langit
- Susan Cline
- Kathy Lass
- Elenor Rieffel
- Elea Chang
- Kaari Peterson
- Shandra Karmarkar
- Samantha Atckins
- Stephanie Moore-Fuller
- Karen McAdams
- Terri Train
- Desi McAdam
- Laura Trippi
- Jenny Greenwood
Notes: by Elea Chang Terminology - network diagrams, cloud represents internet, utilizing cloud, going across any provider w/in limitations. Don't know physically or logically where data is. Unknown
- Lots of discussion to find out cloud taxonomy
- Don't have to provide own processor & storage. Maybe even using someone else's bandwidth
- Networking, computing power, storage --> cloud computing (can rent any of it)**
- Services might build on top. Email's a cloud, UseNet, BitNet
- Cloud computing & SaaS --> SaaS is part of cloud
- National Institute of Science & Technology --> trying to define standards. Other groups protesting
- Cloud providers - who? Amazon, Google, RackSpace, SalesForce (most @ application layer or lower)
- Google application set (mail, docs, sites)
- Trade-offs between cloud computing & own network?
- Companies are not taking responsibility for lost data - "that's your problem"
- How secure is it?
- Being worked on, looking @ it for federal clients...
- Example: my data is running on your infrastructure & federal gov't can ask for it. They can take data and not even tell me. "have to wait 'til court sorts it out"
Hackers --> happens ANYWHERE. That's normal, not specific to cloud computing. Liability is still yours, whether or not it's cloud computing
- Hard to give information to another person unless you give them access @ network level.
When it's good: Large data crunching activities, little websites that are up and very public Cloud doesn't provide you with bandwidth - you buy it. Shifts in cloud computing: Amazon breaks up services to virtual machine, persistent storage. Got a CDN content distribution system
First thing you have to get over: learn whole set of little commands to create instance. Download API toolset, learn how to create instance, kill an instance, manipulate instance. Learning curve -- 'took me about a month'. Had to learn about file system because of snapshot. Had to have disk in a good state to get good look @ system. Have to look @ file system - what can I use to freeze in such a way in a short time that I can do a snapshot. Don't want to have to put up a maintenance page as I back up
Notes: by E. Rieffel
The session was superbly facilitated by Sarah Baker.
Arguments are ongoing as to what cloud computing is, and its relation with software as a service (SaaS), grid computing, utility computing, and distributed computing. Typical uses of cloud computing involve renting computational power, storage, or bandwidth, but some large companies have internal clouds. The key idea is that the user of the cloud does not know exactly which resources will be used; they only request the type and amount of resources they wish to use. A major application is the support of dumb devices, often mobile, that take advantage of CPUs and storage on a cloud. Cloud computing is a new term for something that has been going on in various forms for a long time.
Major cloud providers: Amazon, Google, RackSpace, Sales Force, and Microsoft (Azure). They provide "best effort" security.
Portability and lock in: if you just use the operator system, your stuff will be easy to move. But using the APIs can cause lock in, though your data is generally easy to get out. NIST is developing standard for cloud computing, but these standards are being resisted by cloud providers.
A great service would be if someone could build a meta-cloud in which the user does not have to think about which provider they are using, and portability is automatic.
Amazon and Microsoft use a variant of SQL in their cloud database support.
Sarah Baker gave as an example her experience as the only IT person at a SaaS start up. Using Amazon's cloud reduce monthly costs from approximately $15,000 to $3,000-$5,000. A significant issue is persistence: how to make sure the stuff you need persists past the session. The programmer is responsible for managing persistence. She needed to handle all backups using snapshots, though RackSpace now provides backup support. They were using an all open source infrastructure, including CentOS, a Linux OS with a business spin. Because CentOS does not have a good snapshot/freeze capability, she had the significant challenge of finding her own solution.
Using clouds for e-mail is a problem. All cloud IP addresses are dynamic, so e-mail from these machines will be blocked by spam filters. If you use clouds for e-mail, you must route everything through machines with static IP addresses.
While some of the providers have Hadoop on their clouds, there still is a lot of work to get a good distributed implementation.
Another example: the California election results pages were swamped in the last election, largely because the state did not expect the amount of international interest in the results of proposition 8. Cloud computing is ideal for handling situations like this in which there is a 24 hour spike in the resources needed.