|
Ty Brandenburg
Registered User
Join date: 17 Nov 2007
Posts: 15
|
02-26-2009 18:43
From: Ty Brandenburg Sorry, but you are incorrect. Take a look at VMWare ESX and VMotion. Seamless migration between physical hardware for virtual machines running. Same thing is available for storage systems with storage virtualization. IBM's SVC (SAN Volume Controller) is a good example of that. Move the disk partition (LUN) between physical storage systems while the systems continue to read and write (at fairly high I/Os) without any impact to the system.
Those are two examples that I am intimately familiar with but they are not the only ones. The issue really is one of economics. Does Second Life generate sufficient revenue streams to justify this type of enterprise IT infrastructure.
Regards,
Ty All of that said, updating the underlying Guest OS does require at least a minor reset which would impact the apps running on that OS. Getting SL into a full hitless HA environment would be difficult to say the least. Ty |
|
Sindy Tsure
Will script for shoes
Join date: 18 Sep 2006
Posts: 4,103
|
02-26-2009 19:23
From: Ty Brandenburg All of that said, updating the underlying Guest OS does require at least a minor reset which would impact the apps running on that OS. Getting SL into a full hitless HA environment would be difficult to say the least. Not to mention the expense of OS licenses (vs Debian, which they use now) and, if you really want to be available, FT/HA hardware.. I could see that happening on a small number of critical back-end servers but the whole grid? Not any year soon, I think.. |
|
Gordon Wendt
404 - User not found
Join date: 10 May 2006
Posts: 1,024
|
02-26-2009 20:03
From: Chandra Magic Wait, whut?
So, you're saying they shouldn't be updating the OS on the servers because it's extra work?
Dude, I appreciate how much work it is. I see it very time they're forced to do a rollback because of unseen bugs creeping into the server. I see it every time they release a new viewer RC.
I'm not exactly sure why it is that you're quoting me or anything, unless you think they shouldn't be updating too. In which case, let's migrate the server to Win3.11. It's solid and still working strong today (I know several people that still use it on toy boxes). I didn't mean it as a refute of you, I just quoted you as a point of reference. Of course they should be updating from older programs so that they have as stable and secure a system as possible. I was just commenting that we have it easy as consumer users.
_____________________
Twitter: http://www.twitter.com/GWendt Plurk: http://www.plurk.com/GordonWendt GW Designs: XStreetSL
|
|
Chandra Magic
Registered User
Join date: 12 Jun 2008
Posts: 17
|
02-27-2009 04:01
From: Gordon Wendt I didn't mean it as a refute of you, I just quoted you as a point of reference. Of course they should be updating from older programs so that they have as stable and secure a system as possible. I was just commenting that we have it easy as consumer users. Oh, okei. That works. Had me confused, and yah. We have it much easier by far. |
|
Argent Stonecutter
Emergency Mustelid
Join date: 20 Sep 2005
Posts: 20,263
|
02-27-2009 05:52
From: Ty Brandenburg Sorry, but you are incorrect. Take a look at VMWare ESX and VMotion. Seamless migration between physical hardware for virtual machines running. I support VMWare ESX in my job, and I've worked on safety-critical control systems... stuff where taking too long to recover from a failure can kill people. Seamless migration doesn't mean zero downtime... you still have to snapshot the VM state and restore it on the new server, unless you have the application itself running in a hot standby mode continually synced with the original (which many applications like databases do, but none has as much local state as Second Life). From: Ty Brandenburg All of that said, updating the underlying Guest OS does require at least a minor reset which would impact the apps running on that OS. Getting SL into a full hitless HA environment would be difficult to say the least. Indeed. The amount of work to make SL do hot failover cleanly would make the question of VMs moot... by the time you finished, you'd be able to do it without VMs. |
|
Gordon Wendt
404 - User not found
Join date: 10 May 2006
Posts: 1,024
|
02-27-2009 15:53
From: Argent Stonecutter I support VMWare ESX in my job, and I've worked on safety-critical control systems... stuff where taking too long to recover from a failure can kill people. I knew it, those VMWare installations did it in the library with the candlestick.
_____________________
Twitter: http://www.twitter.com/GWendt Plurk: http://www.plurk.com/GordonWendt GW Designs: XStreetSL
|
|
Prospero Linden
Linden Lab Employee
Join date: 6 Aug 2007
Posts: 315
|
03-03-2009 13:46
Hey all -- the etchification is (finally!) finished , so I'm closing this thread.
OK... there do remain some sim nodes that aren't etchified. However, no more server restarts will be required. "Huh?", you say. Well, we have lots of servers. Lots. This means that we always have some servers that are failed or down. We do regular passes through them and power cycle the ones that aren't behaving well, and RMA the ones that don't come back from a power cycle (or that have needed to be kicked too many times). What this means is that there were some machines that weren't up during this pass, and as they come up we'll still have to etchify them. But, none of you should notice any of this. I'm just kicking that out there in the interest of "gratuitous precision in reporting", and perhaps for a little bit of "too much information."
|
