Facebook may be about to suck up even more of your time. Two weeks ago a feature rolled out that I'm told many users may not have noticed, but that is predicted to soon have them interacting with each other more on the the social network.

That feature was instantly updating comments, and it turns every piece of content on Facebook into an instant messaging chat room. Any time you add a comment to, say, a photo or status update it will now appear instantly in front of anyone else looking at that same page.

It feels like another example of how all kinds of asynchronous messaging tools like email and sms are being converted into synchronous ones. Google Wave (RIP) created a realtime combination of email, IM and wikis; Facebook messages did the same to email, SMS and IM. "Having a conversation in a comment thread was a stuttering experience," Facebook newsfeed engineer Ken Deeter told me, "you had to keep refreshing the page to see new comments."

Switching to live updating may seem a small change from a user's perspective. But the engineering decisions Deeter and colleagues had to make illustrate how complex it is to operate a site with more than half a billion users.

The most obvious way for Facebook to go instant -- having your browser check for new comments every second or so when viewing a page -- "would have made the site fall down for sure," Deeter says. Instead Facebook had be given a kind of short term memory to keep track of who was looking at what at any given moment. Each time a new comment is added it is run past that memory, which checks to see who needs to see it immediately and directs it where it needs to go.

What's more, unlike most data stored in the cloud, it couldn't be put onto hard disks, because that wouldn't keep pace with the rapid rate at which people log on to and off of Facebook. "We're handling around 16 million viewer-to content associations per second so it's unfeasible to do this on disk," says Deeter, "we had to build a memory-based [i.e. RAM] system that can handle a very high write rate."

That new memory couldn't be in one place, because the delay in reaching it from all corners of Facebook's global empire of users would be too great. Instead the team added one in each of the firm's data centers, tasked with keeping track of its closest users. When a comment is added each of the local systems is polled to ensure everyone sees what they need to. You can read more details on this on Facebook's engineering blog.

Any change made by Facebook is a vast social experiment a much as an exercise in software engineering, though, and the firm expects the way people behave on the site to change. In the course of building out the new feature it was uncovered that in any given minute over 100 million comment threads are being viewed online and that over 650,000 comments are made. Those and other figures are now being watched closely. "You have to wait a while for people realize that it is has changed," says Deeter, "in another two weeks or so I think we'll find that people have more conversations in comments because it makes you feel like there are people on the site with you."

To paraphrase, we may all be about to start spending even more time on their site.

Smart Rescheduler can rank possible meeting times. Credit: Google

Google has launched an experimental feature for its Calendar Web application that applies the company's considerable expertise in search to the problem of scheduling meetings. The feature, Smart Rescheduler, allows users to select a meeting that needs to be rescheduled and get suggestions about the best alternative times.

Product manager Ken Norton explains that in the calendar team's surveys of executive administrative assistants, rescheduling was typically a painful, time-consuming task. Norton says that watching administrators "horse-trade" conference rooms, flip through executive calendars, and try to predict the effects of rescheduling a single meeting led the team to believe automation must be able to help somehow.

Google's programmers decided to treat scheduling as a search problem--analogous to finding flights or shopping for products. The Smart Rescheduler takes into account factors such as time zones, available conference rooms, and people who need to attend the meeting, and provides a list of candidate times, ranked by factors such as whether a time is within working hours, whether it is accessible to all attendees, and whether it requires additional rescheduling.

If a conflicting meeting includes many of the same players as the meeting that's being rescheduled, Smart Rescheduler may include the suggestion that the conflicting event could also be rescheduled. A user can also refine the tool by marking certain people as optional or changing the planned length of the meeting. The technology used to rank meeting times is partly borrowed from Google's existing IP and partly built from scratch, Norton says. Smart Rescheduler only works if all participants in a meeting use Google Calendar and if they share availability information with each other. As a result, it's intended mainly for use within a company.

Though it's currently available only as an experimental feature in Calendar Labs, Smart Rescheduler clearly fits into Google's plan to broaden its appeal to enterprise users. In particular, Norton boasted about how the feature demonstrates the power of cloud computing for enterprise. Norton says that the processing required for the ranking algorithms would be too slow if they had to run on a user's local machine--and he didn't miss the opportunity to compare the speed of searching e-mail in Gmail to the speed (or lack thereof) when performing the same search within Outlook.

He also touted Google's ability to release a feature that business users can try just by opting in, without having to upgrade any software.

Amazon has launched a new product that offers up cloud computing possibilities in a model reminiscent of Google AdWords.

Called Spot Instances, it allows users to specify the price they want to pay for access to to resources in the Amazon Elastic Compute Cloud. Users can set other parameters, too, such as the region where they'd like a job to run.

For its part, Amazon gives its unused resources a "spot price" which fluctuates based on how heavily its cloud is being used at a given time. When a user's bid exceeds the spot price, the job runs.

This is obviously a great deal for Amazon, since the company can ensure that it's getting paid for as much of its capacity as possible at any given moment. It's interesting to see the auction model popularized by Google's wildly successful advertising network moving into other types of Internet business.