How to deal with unplanned downtime for users of a Linode se
I was curious if the following event were tot ake place performance-wise and how to deal with it with respect to all users of the services I run:
Suppose unexpectedly, as I've been seeing in the network status forums, that the host my Linode runs on had to reboot or whatever. Even if my Linodes were to come back online, how would I deal with an unplanned downtime? Such a thing would majorly impact performance with my userbase-as I am now moving my IRC server and radio station server etc over to a Linode VPS.
I was curious what you folks thought!
Prior to the move over t o a Linode, I ran thi stuff from an home ISP wich blocked services of all types–so I was violating the rules of the ISP anyways, never was caught at it-though the blocking of port 80 occured most likely because of the trafic I was receiving.
I finally decided because I could not take the time to maintain a bunch of Linux boxes and was often offline and not around to administer those machines in terms of hardware, to switch to Linode. Let the folks at Linode deal with all hardware related stuff-I'd just run the servers.
This has been wonderful, to say the least. I'm just curious if downtime can be a performance-impacting thing, I believe it is thus my choice to post this in this forum.
Any tips etc? Unplanned downtime (unless I cause it myself by say shutting the server off, upgrading/downgrading hosts etc,) is going to be hard to warn users of, that kind of thing. I can't program my IRC client and web software to send notices to all my users of unplanned downtime (especially) if it's on Linode.com's end directly rather than something I've done.
Thanks a lot for all the tips!
Regards, --Keith
3 Replies
One of my Linodes has gone four years with only one outage - a DDoS attack on another user at the same data centre. The other has only been rebooted by Linode twice in two years (the reboots were within a few days of each other while the problem was resolved).
You see a lot of reboots in the System and Network Status forum - but then Linode has a lot of machines (guesstimates run from 500 to 1500 machines).
This works great for connectionless protocols like http, smtp, ntp. It probably wouldn't work great for IRC. But, at least people could reconnect after a minute, rather than however long it takes to fix the host and boot up 40 VPSes.
If you really want to maximize uptime, get two linodes in different DCs (except Atlanta, which blocks IRC and some other stuff).
No idea where you're located. The closest DC to me (in Montreal) is Newark, and after that (ignoring Atlanta), Dallas. If you just want to provide good coverage of the continent, getting both Newark and Fremont is probably going to cover the most population. Fremont for all the western stuff from Vancouver to LA, Newark for the big eastern population centres (NYC, Chicago, Philly, Toronto, Boston, Detroit, Montreal, etc.)