Also, one has to work out the cost of high availability. If you're talking about a situation where you reduce downtime from 4 hours/yr to .5 hours/yr and it costs you x dollars, you have to make sure that the extra 3.5 hours of downtime would cost more than that much money.
The system you have appears to cost $10-100k more than a standard master-slave setup. If you're using a stable unix, stable mysql, good hardware (Sun,HP,IBM), and RAID in a datacenter. You're talking about 99.95% uptime right there. Throw in a slave and it's probably 99.99%. People always seem to forget that downtime is usually caused by human error on a well made system. Human error is what most effort needs to be taken to correct. That means putting your effort into reducing DELETE and UPDATE statement rights, keeping everybody off the machines, having auto_commit off by default. Just think of all the times you've seen a service unavailble due to somebody [EMAIL PROTECTED] up as compared to a kernel fault or a faulty RAID card. > -----Original Message----- > From: Ian Neubert [mailto:[EMAIL PROTECTED] > Sent: Thursday, August 07, 2003 6:23 PM > To: [EMAIL PROTECTED] > Cc: Dathan Vance Pattishall; [EMAIL PROTECTED] > Subject: RE: MySQL Replication > > > Good question :) > > I got a message from a person off the list that suggested I > use network disk > mirroring or a NAS/SAN/NFS system to handle that. I'm not sure if the > mirroring would be 100% perfect, but the NAS/SAN solution > should as either > server would be reading and writing to the same physical data. > > But, then I have another point of failure. Heh. > > I realize that creating the perfect HA system is probably the > most difficult > thing to do, and doesn't come cheaply either. However, I'm > going to think it > through and try anyway :) > > I've read your presentations on your website and have used > that info for my > plan here, but its a little difficult to get details from > just the slides > (as you even mentioned on your site) :) > > Do you bother with multi-masters? How do you ensure redundancy on the > write/master server? > > ....................... > Ian Neubert > Director of IS > TWAcomm.com, Inc. > http://www.twacomm.com/ > > -----Original Message----- > From: Jeremy Zawodny [mailto:[EMAIL PROTECTED] > Sent: Thursday, August 07, 2003 2:53 PM > To: Ian Neubert > Cc: Dathan Vance Pattishall; [EMAIL PROTECTED] > Subject: Re: MySQL Replication > > > On Thu, Aug 07, 2003 at 01:00:12PM -0700, Ian Neubert wrote: > > I was trying to design it so that the slaves wouldn't know they had > > connected to a different master, as they both masters would > have the same > IP > > address that gets failed over based on the Linux Virtual > Server software > and > > VRRP (like heartbeat from Linux-HA). > > That path is a very, very, very difficult one. > > How can you absolutely guarantee that each master's binlog will be > indentical in name, size, and content? > > If you can't, this scenario really falls apart. > > (I've suggested enhancements to MySQL that would fix this but don't > know if they're terribly high on the priority list...) > > Jeremy > -- > Jeremy D. Zawodny | Perl, Web, MySQL, Linux Magazine, Yahoo! > <[EMAIL PROTECTED]> | http://jeremy.zawodny.com/ > > MySQL 4.0.13: up 6 days, processed 212,501,412 queries (399/sec. avg) > > -- > MySQL General Mailing List > For list archives: http://lists.mysql.com/mysql > To unsubscribe: http://lists.mysql.com/[EMAIL PROTECTED] > > > -- MySQL General Mailing List For list archives: http://lists.mysql.com/mysql To unsubscribe: http://lists.mysql.com/[EMAIL PROTECTED]
