Cluster setup

[tnguyen]

I'm very new to H-Sphere and trying to your suggestions for the following cluster setup for reduntancy purposes.

Say I have three servers and I'd like to place each of them at a different data center that way in case something goes wrong with one data center the other two servers can still serve customers.

Is anyone doing this same kind of setup?

If it isn't a good idea to set it up this way, any other recommended ways?

Thanks for your input.

Tom Nguyen

[dynamicnet]

Greetings Tom:

A "general" cluster is that of logical services such as mail, web, database, and such.

While you can run each physical server that runs one or more logical services in different data centers, the only real benefit is that if the data center had data center wide problems that a given set of logical services would not be available.

Therefore, what would be the real value? I am not sure.

Thank you.

[tnguyen]

Well, my idea is to put a cp server in one data center and web/mail on a server at a different data center. Say in case something goes wrong with the data center at which my cp server resides at least my customers can still access their sites and email at the other data center.

Correct me if that isn't how it works.

Thanks,
Tom

[iseletsk]

it will work, but what will happend if second datacenter down, but not first...

[tnguyen]

it will work, but what will happend if second datacenter down, but not first...


What would you recommend? Put all servers at one data center if possible?

Does it speed up connections between servers this way since they're likely on the same network?

Thanks

[iseletsk]

I am just showing that you are not achieving much by doing that.
The main thing is not distributing db & web servers accross datacenters (as there are a lot of traffic between them) - otherwise - you can put mail/cp/dns.. etc in different datacenters with not much of the overhead.
To place or not to place them all into one datacenter is really up to you. From the realibility stand point - it makes sence to have dns in different datacenters.

[tnguyen]

I am just showing that you are not achieving much by doing that.
The main thing is not distributing db & web servers accross datacenters (as there are a lot of traffic between them) - otherwise - you can put mail/cp/dns.. etc in different datacenters with not much of the overhead.
To place or not to place them all into one datacenter is really up to you. From the realibility stand point - it makes sence to have dns in different datacenters.

Is there anyone running everything on one single server that you know of? Any drawbacks by doing this?

I've read Psoft's documentations on moving mail/web to different servers but don't see anything regarding moving DNS. Can DNS be moved to another server? Say I started out with one single server running everything cp/primary dns/secondary dns/database/web, etc and later on add another server. Can I move secondary dns to this second server?

Thanks a lot for your quick response.

[dynamicnet]

Greetings:

We have several customers who run H-Sphere on a single server.

However, I would recommend having a minimum of a two server cluster.

IF you assign each logical service a unique IP address, and IF your data center or IP provider allows you to move those IP's from server to server within their data center, then splitting off logical services to other physical servers is a relative breeze.

Otherwise, it can be done, but there is more hassle.

Thank you.

[tnguyen]

Greetings:

We have several customers who run H-Sphere on a single server.

However, I would recommend having a minimum of a two server cluster.

IF you assign each logical service a unique IP address, and IF your data center or IP provider allows you to move those IP's from server to server within their data center, then splitting off logical services to other physical servers is a relative breeze.

Otherwise, it can be done, but there is more hassle.

Thank you.


Thank you Dynamicnet for your suggestions. I'll consider having two servers instead of one.

Have a nice day.
Tom

[ladylinux]

Hello,

Cp and Tertiary (Third Level) DNS in my Baltimore Datacenter ..

Mail,DNS1 In Texas DC ... (Loaded Web,MySql But Not Set For Sign-Up)

Web1,DNS2,MYSQL in Florida DC (Loaded Mail On This Machine But Not Set For Sign-up)

Backup Server in my local DC ..

Now heres the deal ..

CP dies .. Ok .. mail and web still work ..

Web dies .. cp and mail still work ..

Mail Dies .. cp and web still work ..

If lets say a DC has a issue (Bankruptcy) well you have hot mail and hot web to quickly move your customers from your central backup ..

This was running for 3 months .. until I decided to get out of support business and shut it all down ..

Cost ... About 1,000 a month ..

Customers supported .. Quite a Lot ..

Lady Linux :)

[dorijan]

You dont really need to Spread Accross Several Datacentres.

If you really want to get your uptime up, get in Touch with a Datacentre.
Make a Visit at their place, check out their UPS , overall Security, Environmental Controls. Ask for Remote KVM Solutions, onsite Technicians in case of Problems, Remote Power Bars. Central Backup Servers (i.e Tape Robots)

also get a list of all their possible Carriers, and check out each one of them. Hold some Meetings with them, discuss common Problems such as Routing Issues (what are their plans if a router dies ? or a cable gets cut) or DDoS (how would a DDos on another Customer affect your business) possible UP- and Downgrades. Also Check their Global Routing, what help is a cheap Carrier if your customers get bad Routing and cant access their Service Properly. Also check out Backup Lines.

get in touch with a Hardware Dealer (even try for IBM and Dell , their support contracts are really good)

The last Project I did for a customer was in Switzerland, we have Chosen InterXion Datacentre simply cause they offer us :

99,999% Power Uptime (for the rest we have a Rack Mounted APC UPS)
100% Environmental Protection (even if a plane crashs into the Datacentre our Servers wouldnt have a Problem)
Security (only those who we permit to physically access our servers will be allowed to do so)

We have Chosen a local ISP simply cause of their BGP4 Backbone out of Several bigger Carriers (Level(3), Colt Communications, MFN/Abovenet), their guaranteed uptime of 99,999% for our links.

We also bought a 2Mbit Backup Line from Tiscali International in case the Main Link goes down our Service is still available.

We got our own IP Adresses , simply needed cause if we ever need to switch our ISPs we dont have to change all IPs they stay with us. (ask your ISP for PI - IP Space - Provider Independant IP )

We selected a local Hardware dealer which is about 1 Mile away from our Datacentre, we made a Contract with him to always have enough Spare Parts available for us. We always buy the Same Brand of Machines for our Rack (always the same Motherboard, CPU , RAM , Hard Disk, Network Card) that way we ensure that everything is compatible and has no Problems with the Operating Systems (currently Fedora, FreeBSD, Debian, Windows2003 - check each HCL - Hardware Compatibility List if you are selecting Hardware).

We bought our own Network Equipment :

Cisco Routers simply cause its the preferred branch of all Carriers I have talked to , and they always have a Cisco Certified Technician at the Datacentre who can help in case of Problems.

3Com Managed Switches to built up vLans so Password Sniffing between the Servers of our Customers is not possible. And to get a correct report each Server Customers Transfer Usage.

Plan your Transfer Usage :

Why pay for a 100Mbit line if you only use 5-7Mbits a month. Ask your Carrier if they support bursting. So in case you really need 100Mbit (or more) for one of your Customers you can, but you dont need it 24/7 you might need it during a peek.

Establish Processes for your System Management :

Nothing is more annoying than a Server reboot while your Sales Team is running a Promotion and you expect hundreds of Users accessing your Website and they cant reach it.

if you have more questions simply ask.

[ladylinux]

You dont really need to Spread Accross Several Datacentres.

If you really want to get your uptime up, get in Touch with a Datacentre.
Make a Visit at their place, check out their UPS , overall Security, Environmental Controls. Ask for Remote KVM Solutions, onsite Technicians in case of Problems, Remote Power Bars. Central Backup Servers (i.e Tape Robots)

I speaking to the people who buy servers at 100 a month ..

Which would be most the folks here ..

You are 1000% right though ..

Thats why I have my own DC here in Balto ..

Lady Linux

[dorijan]

The Datacentre my Cluster is in offers a Setup like that.

Standard Servers are

Celeron 2,6GHZ 512RAM 60GB HDD - FreeBSD / Debian / SuSE [Redhat EL costs extra +15 Euro / Month] with 1000GB Xfer - 100 Euro/ Month

if you have more than one , the bandwidth is pooled for all your servers. (i.e buy 3 servers you get 3TB Xfer and they dont care how you spent it)

Included with each Server is Backup to a Central Backup Server, you get 5GB Backup Transfer for each Server - (you can store unlimited data they only charge the actual bandwidth to the server)

You can aswell buy a so called Dedicated System, which gives you a Private Switch and a better Backbone Connect. Also internal Bandwidth between Servers isnt counted anymore. + it upgrades your uptime guarantee to 99,99%

costs around 150 Euro a Month.

That leaves you with about 200 Euro / Month for a Standard Cluster of 2 Servers and FreeBSD with 99,9% Uptime, 4h Hardware Guarantee and 24/7 Toll Free Phone Support. And you dont have to worry about any Power Stuff, or Hardware or Backbone/Routing issues. If your System is not reachable 99,9% you get a Certain ammount of Money back.

I dont want to advertise them on here thats why names are kept back.

[ladylinux]

Celeron 2,6GHZ 512RAM 60GB HDD - FreeBSD / Debian / SuSE [Redhat EL costs extra +15 Euro / Month] with 1000GB Xfer - 100 Euro/ Month

Hello,

Ok ..well you were throwing in stuff like tape robots .. private lines ..

Your own Cisco equipment .. etc etc etc ..

Thats gonna get you reliability ..

But if you are not near a DC .. and dont have the cash for all that ..

You are at the mercy of 100 a month leased servers ..

Good solutions both of them ..

One (Yours) is tightly under your control .. and does cost .. and is awesome .. :-)

Mine Is loosely spread about .. somewhat under my control .. (Years of experience and Im not gonna touch that plane thing and servers staying up .. have told me .. you can plan plan plan .. and a lil mouse of a bug will find a way to muck it all up in a second.)

Anyhow .. I feel as though this guy might be starting out and our discussion has strayed this thread badly .. :-)

Lady Linux

[dorijan]

Im doing consulting and so far most of the jobs i did for Webhosting Companies was to "rescue" them off the Multi Datacentre, Multi Control Panel Solution into one Solution in one Place , some had the Money to buy their own Dataroom, others just went with a Solution most European Based Datacentres offer. (Im not familiar with the US so i dont dare to rate them at all) I just thought our little discussion and the info inside can help the thread starter to decide better what he really needs and how much he can afford.

Thanks

Regards
Christian

[ladylinux]

Im doing consulting and so far most of the jobs i did for Webhosting Companies was to "rescue" them off the Multi Datacentre, Multi Control Panel Solution into one Solution in one Place

I used to work with Jet Propulsion Laboratorys .. Before IP was even used we had a distributed system of communication called SPAN .. Then came the Internet .. And maybe I keep going back to why the Internet was set up in the first place .. To survive nuclear war ..

LOL .. its hardly that anymore ..

And my design principles maybe reflect the idea that if one goes down another at least heartbeats info to the customer ..

But thank you .. The conversation while boring to most is interesting to me ..

Lady Linux :)

[tnguyen]

Thanks for everyone's input. I will definitely take your input into consideration when the time comes as I will only start out with just one single server for right now and might not expand more until next year.

Thanks again.
Tom