Consistency, Availability, Reliability

Post #274,149 by tuberculosis 11/27/06 11:21:47 PM Reply	Consistency, Availability, Reliability Pick any 2. We can go from there. [link\|http://www.blackbagops.net\|Black Bag Operations Log] [link\|http://www.objectiveclips.com\|Artificial Intelligence] [link\|http://www.badpage.info/seaside/html\|Scrutinizer]
Post #274,151 by admin 11/27/06 11:34:27 PM Reply	Define your terms. Or define reliability at least. I'd be inclined to favor the first two. Consistency is ultra-important. Server lag is a reasonably close second. How important reliability is depends on what you mean to give up; that kind of choice is usually more of a continuum than boolean. Regards, -scott anderson "Welcome to Rivendell, Mr. Anderson..."
Post #274,163 by tuberculosis 11/28/06 12:29:13 AM Reply	mmm kay They are all a continuum, but generally you can only get two at a time. Consistency means that whatever data you have is never stale. It is never wrong. You relax this by saying it is never more than one version out of date or it is never more than 2 minutes old. Availability means you always have data at your fingertips. You never have to wait long for it. Fetches return immediately with what you expect them to. The data may or may not be current. You relax this by allowing things like "Data is locked by other user - please come back tomorrow" states. Reliability or Scalability implies that the system is always up - it can scale with just the addition of hardware and is capable of withstanding component/peer failure. Relaxing this means using something like a central lock server to manage data consistency introduces a single point of failure that will take down your system. Doing this will cause to to relax availability. Here's a great talk about this stuff by Werner Vogels. [link\|http://www.itconversations.com/shows/detail459.html\|http://www.itconvers...ws/detail459.html] [link\|http://www.blackbagops.net\|Black Bag Operations Log] [link\|http://www.objectiveclips.com\|Artificial Intelligence] [link\|http://www.badpage.info/seaside/html\|Scrutinizer]
Post #274,208 by admin 11/28/06 12:09:15 PM Reply	That's what I thought, for the most part. So consistency, then availability, then reliability. One process per instance, reading all of the data into RAM, then with lazy writes to persistent storage fits the above, I think. Enough of the scalability can still be there, I think, for purposes of the example problem we're discussing, since each instance gets its own process. Werner's blog won't load right now; I'll check it out again later. Regards, -scott anderson "Welcome to Rivendell, Mr. Anderson..."
Post #274,210 by admin 11/28/06 12:12:41 PM Reply	Or how about we do this: Let's talk about the solutions three ways, one for each of the tripod being dropped. Regards, -scott anderson "Welcome to Rivendell, Mr. Anderson..."

Welcome to IWETHEY!