| Author |
Topic |
|
SamC
White Water Yakist
3467 Posts |
 Posted - 2003-11-13 : 13:20:53
|
| No Single Point Of FailureWhat are the cost effective solution(s) to migrate a stand alone server which hosts IIS and SQL into a new hardware / configuration that would not have any single point of failure?Sam |
|
|
tkizer
Almighty SQL Goddess
38200 Posts |
Posted - 2003-11-13 : 13:24:28
|
| Any single point of failure? We're talking about very expensive "High Availability" hardware and software. MS has great documentation about their HA solution. It involves clustering with Windows Data Center as the OS. Is this what you are referring to? We're talking 6 zeroes in the price tag though.Tara |
 |
|
|
SamC
White Water Yakist
3467 Posts |
Posted - 2003-11-13 : 14:22:20
|
| I'm sure there are 6 zero solutions that I can't afford. I was hoping there might be a solution involving 4 (or fewer) inexpensive fault-intolerant servers. Not having any practical experience here I thought there may be someone on SQLTEAM who had experimented with different server redundancy configurations.How about two .NET servers front ending a two clustered SQL servers ?? I've read .NET can be setup this way so a single failure doesn't kill the web, and *I think* that the session context can be shared among all the .NET servers. I don't know about clustered server functionality/pricing on two fault intolerant servers.Sam |
|
|
|
MichaelP
Jedi Yak
2489 Posts |
Posted - 2003-11-13 : 15:04:26
|
| Well, you need a minumim of four servers. If you really want no single point, you also need 2 switches, redundant power supplies in each server, UPS's, etc.You'll need to setup an Active / Passice Cluster for your SQL servers. That means one server is doing all the work, the other is doing nothing. You can setup some active / active stuff, but that probably doesn't aply here. On the webservers, are you running ASP 3.0 or ASP.net? For 3.0, i wrote some custom session stuff that stored state in the database, and that made everything work since all the webservers in my server farm could get the state out. It didn't matter what webserver a user hit from page to page. For ASP.net it has a built in state management system that can use SQL server to hold the state. Very handy. To balance between your two web servers, you should probably get a hardware load balancer. We use a Cisco Local Director. It's not the easist or cheapest thing to use, but it works great. You could also use the built in MS Network Load Balancing in Windows 2000 (might be called something else in 2003 not sure). It bacially balances web requests between N servers, and it's free! Let me know if you have any more questions about this. I've had a system like this running for several years for way under six zero's.Michael<Yoda>Use the Search page you must. Find the answer you will.</Yoda> |
|
|
|
tkizer
Almighty SQL Goddess
38200 Posts |
Posted - 2003-11-13 : 15:06:02
|
| Six zeroes was in reference to an HA environment with 5 9s of reliability. Yes you can have a different environment that costs less.Tara |
|
|
|
MichaelP
Jedi Yak
2489 Posts |
Posted - 2003-11-13 : 15:06:51
|
| Oh, as far as finding inexpensive servers, you can find some great deals on Ebay. Depending on your hardware requirements, you could pickup a few servers for under $2000 that will do what you need.Michael<Yoda>Use the Search page you must. Find the answer you will.</Yoda> |
|
|
|
MichaelP
Jedi Yak
2489 Posts |
Posted - 2003-11-13 : 15:10:05
|
| Oh, I see what you are saying Tara. Yeah, if you want true 99.99999% uptime, it's going to cost a LOT. If you want something more like 99.99% or so uptime, my solutions will do it for ya. It comes down to how mcuh do you have to spend. The more you can spend, the more reliable the system can become. You also haev to look at the cost of downtime. If a few hours of downtime will cost you tens of thousands of dollars, it's MUCH cheaper to build a bad ass NSPOF system up front.At the risk of sounding like P-Diddy, it's all about the Benjamins!Michael<Yoda>Use the Search page you must. Find the answer you will.</Yoda> |
|
|
|
SamC
White Water Yakist
3467 Posts |
Posted - 2003-11-13 : 15:18:18
|
| Tara - sounds like the HA system you have doesn't even burp if a CPU or memory chip fails. In that environment, as long as you have backup, everythings covered.My goal was to setup a system that would keep running while the failed component was repaired or replaced. Then there's the issue of bringing it back into operation and carrying part of the load, ensuring the hard drives mirror each other properly and on and on.I wonder if HA provides fewer things to worry about by taking care of more issues that must be manually solved in Michael's configuration?Though I am familiar with it, I haven't taken care of the ASP state in a database thing. I'd probably migrate the last ASP to .NET and let it take care of state.How does an Active / Passive SQL server keep the database content up to date on the passive server? Sounds like some data would be lost?Sam |
|
|
|
tkizer
Almighty SQL Goddess
38200 Posts |
Posted - 2003-11-13 : 15:20:18
|
| Sam, the other server doesn't have the drives until you switch SQL Server over to it. Whichever server has SQL Server running on it currently has the drives. No data is lost because it is stored in one place.Tara |
|
|
|
tkizer
Almighty SQL Goddess
38200 Posts |
Posted - 2003-11-13 : 15:22:31
|
| When we need to perform hardware maintenance or any kind of maintenance, we move SQL Server over to the other server. This takes about 30 seconds to complete. So there is 30 seconds of downtime where nothing can reach it as it transfers control to the other server. We then complete the maintenance and might leave SQL Server on the other server if we want. In our environment though, we have an active/active cluster. So we do move SQL Server back to the original server so that the servers are balanced with databases.Tara |
|
|
|
MichaelP
Jedi Yak
2489 Posts |
Posted - 2003-11-13 : 15:27:48
|
| Sam, in both Tara's suggestion's and mine, if some hardware fails on the Cluster, then the Cluster automatically fails over to the Passive node, and operations continue. There's about a 30 second downtime, but customers never seems to notice that on our web apps. The reason that this works is because the two cluster boxes have a shared storage device where all of your SQL data is stored. Michael<Yoda>Use the Search page you must. Find the answer you will.</Yoda> |
|
|
|
Blastrix
Posting Yak Master
208 Posts |
Posted - 2003-11-13 : 15:50:32
|
| Tara and Michael,You both mentioned that the switch to the passive node only takes about 30 seconds. I don't doubt you, but why do I hear from many people and newsgroups and such that if the active server chokes, then it can 5 minutes or more for the new server to take over and do what it needs to do? Are these people misinformed or is the situation different when the active server dies rather than being gracefully switched over?Also, there's no way to actually balance load with a cluster setup, correct? The only way to sustain more load is to upgrade the machines being used in the cluster, right?Steve |
|
|
|
X002548
Not Just a Number
15586 Posts |
Posted - 2003-11-13 : 15:58:57
|
quote:
Originally posted by Blastrix
there's no way to actually balance load with a cluster setup, correct? The only way to sustain more load is to upgrade the machines
Yup...BIG misconception....MS gave up on scaling out...only way is up....Now Oracle....look at for "Grid" computing...True scaling out...kinda like a mainframe...Brett8-) |
|
|
|
MichaelP
Jedi Yak
2489 Posts |
Posted - 2003-11-13 : 16:18:12
|
| Steve, In their environments, it might take as much as five minutes, but in my environment, I've never seen it take more than 90 seconds. It usually only takes about 30 seconds. Microsoft clustering is for redundancy, and not load balancing. You must scale up just liek Brett says. Scaling up is making your two clustered PC's bigger and better by adding more CPU/ RAM to them. I highly suggest getting a 4-8way capable box when building your cluster up Sam. Even if you only need 2 CPU's now, when you need that 4th CPU, you'll love not having to rebuilt your cluster. Buy hardware big enough that will last you about 3 years or so. We started at 2 CPU's an now we are up to 6 CPU's. Thank god for our 6-way server, or we would have had to rebuilt the entire thing (not fun). Oh, that brings me to another point. Setting up MS clustering is NOT FUN or EASY. You must get hardware and drivers that support MS clustering, or you'll never get it working reliably. One easy solution is a "cluster in a box" that gives you everything you need to setup a cluster. I know HP/Compaw has such a solution.MichaelMichael<Yoda>Use the Search page you must. Find the answer you will.</Yoda> |
|
|
|
SamC
White Water Yakist
3467 Posts |
Posted - 2003-11-13 : 16:29:00
|
| Thanks for everyone's contribution. I learned a lot. Cluster in a box. Switchover time.If we get a go-ahead to do this in the coming months, I'll have further posts on this topic.Sam |
|
|
|
gercr
Yak Posting Veteran
53 Posts |
Posted - 2003-11-13 : 16:57:28
|
| By the way, Sam you must check if your applications are cluster aware, this isn't necessary but is best! |
|
|
|
Merkin
Funky Drop Bear Fearing SQL Dude!
4970 Posts |
Posted - 2003-11-13 : 17:13:48
|
quote:
Originally posted by tdugganNo data is lost because it is stored in one place.
Which is a single point of failure (although fairly unlikely).I had a fibre hub die on me. That will put a fairly big hole in your cluster if the servers can't see the shared drives.Damian |
|
|
|
tkizer
Almighty SQL Goddess
38200 Posts |
Posted - 2003-11-13 : 17:18:52
|
| Hmmm, I wonder how we have it configured then, because we have no single point of failure from what I understand. They spent about 6 months building the environment with the support of the vendor. I'll have to find out.Tara |
|
|
|
Merkin
Funky Drop Bear Fearing SQL Dude!
4970 Posts |
Posted - 2003-11-13 : 17:20:38
|
Oh I'm sure your budget is MUCH bigger than the one I was working on  Damian |
|
|
|
MichaelP
Jedi Yak
2489 Posts |
Posted - 2003-11-13 : 17:21:55
|
| Damian, that's why you haev 2 fibre hubs, and 2 HBA's in each box. Redundant paths to the shared storage :)It's ain't cheap, but it doesn't fail.Michael<Yoda>Use the Search page you must. Find the answer you will.</Yoda> |
|
|
|
Merkin
Funky Drop Bear Fearing SQL Dude!
4970 Posts |
Posted - 2003-11-13 : 17:25:37
|
quote:
Originally posted by MichaelP
Damian, that's why you haev 2 fibre hubs, and 2 HBA's in each box. Redundant paths to the shared storage :)
That's what I thought the answer would be. Thanks Michael.Damian |
|
|
|
Next Page
|
NSPOF