Can I use normal computer parts for a server

[Zubbie]

Hello, I'm building a server for a business website (I'm coding my own). Am I able to use normal computer parts (i.e. I7 6700) for the server? Thanks

 

[kanewolf]

You may want to use ECC (error correcting) RAM and find a chassis with dual power supplies. Power supply failure is the most common cause for down time.
Other "server" features that can be useful are IPMI (dedicated management network interface), hot swap disk drives, etc.

It really depends on how "expensive" an outage is.

 

[the nerd 389]

Servers don't normally use desktop parts for some pretty specific reasons. If those reasons aren't important to you, go for it.

The first reason to use server parts is reliability. Server parts undergo more stringent reliability testing. They also support additional reliability features like ECC memory.

The second reason is that server parts are designed to be fully loaded 24/7. This means that they are supposed to deliver consistent performance rather than kicking into a turbo mode like computer parts. This allows server parts to have lower specifications than consumer parts would need to deliver the same performance, as the server part generally has full performance on tap at any time. Consumer parts use many strategies to extract more short-term performance at the expense of steady state performance.

Third, server parts are designed to be stable. That means that they tend to be more resilient to adverse operating conditions, like power surges, dirty power, etc. This does not apply to cooling, though. Server parts normally assume that ample cooling headroom will be available.

Lastly, server parts are generally verified to work with a different class of software than consumer parts. This means that the authors of hosting software are more likely to test the software on server hardware than they are on consumer hardware.

For a basic server like the one you're talking about, you can probably use consumer parts for most components without issue, but some should be server-grade. The main items that come to mind are the storage drives and the PSU. Also, the i7-6700 is total overkill for basic web hosting.

 

[Zubbie]

Servers don't normally use desktop parts for some pretty specific reasons. If those reasons aren't important to you, go for it.

The first reason to use server parts is reliability. Server parts undergo more stringent reliability testing. They also support additional reliability features like ECC memory.

The second reason is that server parts are designed to be fully loaded 24/7. This means that they are supposed to deliver consistent performance rather than kicking into a turbo mode like computer parts. This allows server parts to have lower specifications than consumer parts would need to deliver the same performance, as the server part generally has full performance on tap at any time. Consumer parts use many strategies to extract more short-term performance at the expense of steady state performance.

Third, server parts are designed to be stable. That means that they tend to be more resilient to adverse operating conditions, like power surges, dirty power, etc. This does not apply to cooling, though. Server parts normally assume that ample cooling headroom will be available.

Lastly, server parts are generally verified to work with a different class of software than consumer parts. This means that the authors of hosting software are more likely to test the software on server hardware than they are on consumer hardware.

For a basic server like the one you're talking about, you can probably use consumer parts for most components without issue, but some should be server-grade. The main items that come to mind are the storage drives and the PSU. Also, the i7-6700 is total overkill for basic web hosting.

So should I use like a core i3 or go with a third generation Xeon? Also would I need raid controllers?

 

[kanewolf]

What is the cost of downtime compared to the cost of the more reliable parts? That is the question you have to answer. Is $1000 extra on hardware going to provide a benefit because a 6 hour outage would cost the customer $10K ? Would you be better off with multiple cheaper servers and a network load balancer to provide redundancy?

I might say you would be best served by not self-hosting. Purchase a hosting service and use their computers and bandwidth.

 

[the nerd 389]

Personally, I would use a hosting service for 95% of the websites that I deal with. The costs, reliability, and maintenance concerns all push in that direction. The only real advantage of on-site web hosting is better control. kanewolf's comment is spot-on regarding how to decide between the two.

If you're able to tell us, more information regarding the type of website you want to host would help us make better recommendations.

If you just want to understand the thought process well enough to make the judgement yourself, I'll try to condense it down enough to be digestible. There are three pieces to the puzzle. First, you have to gauge the cost of downtime. This will guide you when choosing hardware necessary to get acceptable performance. This has a massive impact in the decisions made in the other parts of the puzzle.

The second part of the puzzle is determining the requirements of the server. These include the reliability, performance, and TCO that you need to get from the server. For reliability, you normally want something between three 9's (99.9% up time) and five 9's (99.999% up time) of reliability. Sometimes, you need more, but that's usually in very specific cases. Three 9's of reliability means you can have the server down for up to 8 hours per year. If it's only one server, that means you can generally only reboot once per week, and you'd have one chance per year to make a hardware change. When considering this number, you have to consider the possibility of hardware failure, the internet connection reliability, the occasional need to update the server, the impact of specific components failing, and the power quality in your area. In my experience, hot-swap hard drives (which means you need a RAID card) and a redundant PSU are requirements for achieving three 9's of reliability. Achieving any more than that is nearly impossible for most small businesses, as it would require backup power for the server and internet connection, and possibly a redundant server.

The amount of processing power you need in a server depends on exactly what you're trying to do with it. If you're running a basic website with no database functions, and little server side scripting, you can get perfectly acceptable performance from 2 generation old Celerons. In fact, those are capable enough to run a Minecraft server for several players. The field changes when you start adding in database functionality, heavy server side scripting, and other features. Aside from compute servers and virtualization, you normally only need a Xeon E3-1240 at most for a small-business website (I'm assuming that the server will only have one function, as is best practice), and even that is a rare thing indeed.

For TCO, you have to consider the lifetime and cost of each hardware component, the cost of electricity (including cooling), the cost of a suitable internet connection, and the cost of the time spent maintaining the system. Honestly, I've never really dealt with this specific metric much, with the exception of time spend maintaining the system and the cost/lifetime of components.

At this point, you're ready for the third piece: to look into what components are suitable for your server, and then compare costs with cloud alternatives. I've found that in general, cloud services are the better deal (often by a wide margin). If you need the control of on-site hosting, then be prepared to pay a pretty penny up front as well as for the internet connection.

 

[Zubbie]

Personally, I would use a hosting service for 95% of the websites that I deal with. The costs, reliability, and maintenance concerns all push in that direction. The only real advantage of on-site web hosting is better control. kanewolf's comment is spot-on regarding how to decide between the two.

If you're able to tell us, more information regarding the type of website you want to host would help us make better recommendations.

If you just want to understand the thought process well enough to make the judgement yourself, I'll try to condense it down enough to be digestible. There are three pieces to the puzzle. First, you have to gauge the cost of downtime. This will guide you when choosing hardware necessary to get acceptable performance. This has a massive impact in the decisions made in the other parts of the puzzle.

The second part of the puzzle is determining the requirements of the server. These include the reliability, performance, and TCO that you need to get from the server. For reliability, you normally want something between three 9's (99.9% up time) and five 9's (99.999% up time) of reliability. Sometimes, you need more, but that's usually in very specific cases. Three 9's of reliability means you can have the server down for up to 8 hours per year. If it's only one server, that means you can generally only reboot once per week, and you'd have one chance per year to make a hardware change. When considering this number, you have to consider the possibility of hardware failure, the internet connection reliability, the occasional need to update the server, the impact of specific components failing, and the power quality in your area. In my experience, hot-swap hard drives (which means you need a RAID card) and a redundant PSU are requirements for achieving three 9's of reliability. Achieving any more than that is nearly impossible for most small businesses, as it would require backup power for the server and internet connection, and possibly a redundant server.

The amount of processing power you need in a server depends on exactly what you're trying to do with it. If you're running a basic website with no database functions, and little server side scripting, you can get perfectly acceptable performance from 2 generation old Celerons. In fact, those are capable enough to run a Minecraft server for several players. The field changes when you start adding in database functionality, heavy server side scripting, and other features. Aside from compute servers and virtualization, you normally only need a Xeon E3-1240 at most for a small-business website (I'm assuming that the server will only have one function, as is best practice), and even that is a rare thing indeed.

For TCO, you have to consider the lifetime and cost of each hardware component, the cost of electricity (including cooling), the cost of a suitable internet connection, and the cost of the time spent maintaining the system. Honestly, I've never really dealt with this specific metric much, with the exception of time spend maintaining the system and the cost/lifetime of components.

At this point, you're ready for the third piece: to look into what components are suitable for your server, and then compare costs with cloud alternatives. I've found that in general, cloud services are the better deal (often by a wide margin). If you need the control of on-site hosting, then be prepared to pay a pretty penny up front as well as for the internet connection.

The website is very basic with information on products (soon to be more) and information on the company. It will also have a phone numbers and a link to Facebook.

 

[kanewolf]

I would HIGHLY recommend a hosting service. They can amortize the costs associated with high quality network security among the 100s or 1000s of subscribers. Self hosting opens you up to unending attacks. Let them fight those off.

 

[the nerd 389]

Is there a specific reason you need to have the server on-site?

If not, put it in the cloud. It'll be a lot cheaper, and at least as reliable as anything you'd put together.

To get the same reliability as the cloud, you're looking at roughly $5000-$10000 between the time and hardware you'll need. That doesn't include software, as those prices vary way too much to give a simple estimate.

If you just want a simple, temporary build, you could get away with a Sandy Bridge Celeron, perhaps less. You would want to have a pair of small drives in RAID 1 for the site and a separate drive for the OS, and a redundant PSU with UPS backup. For RAM, the website is trivial. The OS will use more RAM than the website. For the internet, get a business connection with at least three 9's of uptime. Almost any business class router will do, but if the connection will be shared with the office, it should have DMZ capabilities. For the chassis, make sure it has at least two fans.

Alternatively, you could get an entry level server with those features from Dell, HP, or any other OEM of your choice.

Finally, make sure you have a decent backup system in place.