need econo webserver for Win2k3, IIS, CFM

[TechBoyJK]

I need to buy/build 4 Win2k3 webservers to run IIS and CFM (bluedragon). I also need to build 2 Win2k3 SQL 2000 servers, with one to be the primary server, and the other as a backup (using transactional replication).

My question is, what is the ideal system and a quality per cost ratio? I was thinking of building the system's myself. For instance, in the case of the webservers, find a nice 3U Antec Case with rails, buy a good dual core or dual cpu motherboard, get 2GB of RAM, and use the MB's onboard sata raid controller to mirror two 100+GB sata 10k drives. I think with a good chipset (like Core 2 DUO or AMD Opteron), these should provide excellent performance.

I already have rack space in a datacenter, power, and internet. So thats not an issue. I just want to know the best way to acquire 4 cost effective webservers that will be reliable.


Webservers:

I want to load balance them, with 3 in active and 1 in standby. I'm going to use direct routing, in that as the load balancer distributes connections, the client will deal directly with a webserver once it is directed there. This prevents return traffic from having to funnel back through the Load Balancer. So yea, 3 active, 1 in standby.

I can build these myself, but if I can get a quality setup from Dell in a nice case and on rails, then I might do that instead.

I don't think each server will need more than 2GB of RAM or dual cpu's. If load gets greater than what the setup can support, I can always add another machine, and it could be the start of an beefier array if funds allow. Keeping at 2GB of RAM and dual cpu's (or dual core), I can get away with Windows 2003 Webserver Edition, which aligns with the server's purpose.

Database Servers:

I am thinking Windows 2K3 Standard, with a dual, dual core cpu's and 4GB of ram. I'd probably use seperate disks for OS, Logs, and DATA, so I would probably put like raid 1 mirrored 60GB sata drives for the OS, a single 500GB drive for logs, and raid 1 mirrored 120GB sata drives for data. I assume for this configuration, I'll probably need an additional sata raid controller for the second raid array.


SO I ask, what are my best options here. I'd like to spend no more than $800 per webserver and $1500 per DB server.

 

[TechBoyJK]

bmp

 

[RebateMonger]

I'm definitely not an expert in Load Balancing and Clustering. But here's my understanding of some of the basic requirements for Network Load Balancing Clusters and Server Clusters in the Server 2003 world:

"Network Load Balancing Clusters" are suitable for providing high availability, high reliability, and scalable services that are NOT stateful. Examples of mostly non-stateful applications include web servers, FTP servers, and VPN servers. The data itself rarely changes. Typically, each server has its own data storage, which can be of any type.

The NLB Cluster can be built with these Server 2003 Editions: Web, Standard, Enterprise, and Datacenter. All Servers are Active and listen to requests for service. Each Server decides if it will accept a service request.

"Server Clusters" are suitable for for providing high availability services that ARE stateful. Examples of stateful applications include database servers, email servers, and file and print servers. The data itself is frequently changing. Server Clusters share a single data storage device. Historically, Server Clusters only supported SCSI arrays (maximum 2-node Cluster) or Fibre-Channel arrays, and, (I believe) iSCSI arrays and SAS arrays.

The Server Cluster can be built with these Server 2003 Editions: Enterprise and Datacenter. Each Server can be designated Active or Passive.

Microsoft's recommendations on using Replication, Log Shipping, or Clustering as a Failover Solution:

"Should I Use Replication, Log Shipping, or Clustering as a Failover Solution?

? Use SQL Server clustering when you want a hot, standby, failover support. Clustering offers automatic detection of server failure and automatic failover between nodes participating in the cluster. Client application connections can automatically be redirected to the current active server instance and application data requests will be serviced with little or no interruption.

? Use log shipping when you want a warm standby solution. Log shipping transfers database dumps and logs from the primary server to the secondary server. As with any warm standby solution, failure detection and failover processing is not automated. When log shipping is employed, the standby database cannot be used for data access. The standby database is unavailable throughout the continual process of loading database and transaction log dumps from the primary server. Because log shipping transfers complete database logs, all logged schema and data changes are transferred from the primary to secondary node. When log shipping is employed, it is not possible to partition the data or transfer only a subset of the objects from the database.

? Use replication when you want a warm standby solution, and when simultaneous data access on the secondary node (scale out for read) or object or data partitioning is a requirement.

Clustering is a shared disk, hardware solution protecting against computer failure. The cluster computers are usually in close proximity to one another. Log shipping and replication provide geographically disparate copies of the database. A combination of these solutions can be used to provide protection against different types of failure."

 

[Sunner]

Well, sizing up servers for Windows is definitely not in my area of expertise, as I'm more of a Solaris/Linux kind of guy

However, as a more general piece of advice, if you are getting them for a proper business, which it certainly sounds like, DO NOT BUILD THEM YOURSELF.
For one, a prebuilt from a good OEM(I like HP's ProLiants myself) will have a lot of engineering in it, in terms of proper cooling, reliability, cabling, and compatibility, a home built will never match that.
Secondly, you will have one source for everything, with a basic support contract that you can upgrade should you see the need.
Aside from such things as having a guaranteed period of spare parts, this is also a place to point your finger at when the ****** hits the fan, which, depending on where you work, might very well save your ass from the wrath of your boss.

Besides, lower end servers experience pretty much the same cut throat competition as consumer hardware, so they're really not that much more than a home built, even less sometimes.
A ProLiant DL380, which is a very nice all around server, can run as low as $1200 or thereabouts if you find a deal on them.
That's minus disks of course, but still, it's not exactly expensive for such a high quality piece of hardware.

Besides, if you decide to quit, your replacement will love you for buying from a good vendor, he'll know who to call in case of trouble just by looking at the label, and the person on the other end of the line will know what spare parts to send, etc etc.

 

[TechBoyJK]

Originally posted by: RebateMonger
I'm definitely not an expert in Load Balancing and Clustering. But here's my understanding of some of the basic requirements for Network Load Balancing Clusters and Server Clusters in the Server 2003 world:

"Network Load Balancing Clusters" are suitable for providing high availability, high reliability, and scalable services that are NOT stateful. Examples of mostly non-stateful applications include web servers, FTP servers, and VPN servers. The data itself rarely changes. Typically, each server has its own data storage, which can be of any type.

The NLB Cluster can be built with these Server 2003 Editions: Web, Standard, Enterprise, and Datacenter. All Servers are Active and listen to requests for service. Each Server decides if it will accept a service request.

"Server Clusters" are suitable for for providing high availability services that ARE stateful. Examples of stateful applications include database servers, email servers, and file and print servers. The data itself is frequently changing. Server Clusters share a single data storage device. Historically, Server Clusters only supported SCSI arrays (maximum 2-node Cluster) or Fibre-Channel arrays, and, (I believe) iSCSI arrays and SAS arrays.

The Server Cluster can be built with these Server 2003 Editions: Enterprise and Datacenter. Each Server can be designated Active or Passive.

Microsoft's recommendations on using Replication, Log Shipping, or Clustering as a Failover Solution:

"Should I Use Replication, Log Shipping, or Clustering as a Failover Solution?

? Use SQL Server clustering when you want a hot, standby, failover support. Clustering offers automatic detection of server failure and automatic failover between nodes participating in the cluster. Client application connections can automatically be redirected to the current active server instance and application data requests will be serviced with little or no interruption.

? Use log shipping when you want a warm standby solution. Log shipping transfers database dumps and logs from the primary server to the secondary server. As with any warm standby solution, failure detection and failover processing is not automated. When log shipping is employed, the standby database cannot be used for data access. The standby database is unavailable throughout the continual process of loading database and transaction log dumps from the primary server. Because log shipping transfers complete database logs, all logged schema and data changes are transferred from the primary to secondary node. When log shipping is employed, it is not possible to partition the data or transfer only a subset of the objects from the database.

? Use replication when you want a warm standby solution, and when simultaneous data access on the secondary node (scale out for read) or object or data partitioning is a requirement.

Clustering is a shared disk, hardware solution protecting against computer failure. The cluster computers are usually in close proximity to one another. Log shipping and replication provide geographically disparate copies of the database. A combination of these solutions can be used to provide protection against different types of failure."

Thanks. Im aware of all this as I am certified in HA Solutions, which was my elective for MSCE. I'm actually performing the NLB via an F5, so the webservers are sandboxed from each other and don't need any application or OS level load balancing.

SQL servers are not so easy...... =)