Soldato
- Joined
- 7 Jul 2010
- Posts
- 3,581
Just use server hardware
Custom Servers
- Thread starter anything I don't mind
- Start date
A wise old man once told me.
"If you build the client a server, it goes bang and you can't fix it quickly, your in trouble. If the customer buys an IBM/Dell/HP server, it goes bang and the manufacturer can't fix it quickly. They are in trouble".
"If you build the client a server, it goes bang and you can't fix it quickly, your in trouble. If the customer buys an IBM/Dell/HP server, it goes bang and the manufacturer can't fix it quickly. They are in trouble".
We use a couple custom built VM hosts running Hyper-V. We built them for around £1200 ex vat each. As soon as you mention >24GB RAM to Dell/HP they are immediately sucking air through their teeth. The prices become utterly ridiculous for a small business.
I don't really see the issue. If you know what you are doing and know the quality of hardware you are putting together then go for it. We stuck to all 100% Intel chipset kit just to minimize any niggly driver problems.
Currently we're running single PSUs in them but there's nothing stopping us upgrading the PSUs to some dual redundant ones at a later date.
We came close to going with high-end desktop hardware. But in the end we opted for a decent server/workstation class motherboard (think along lines of Supermicro or Tyan). The key thing is to choose the right motherboard.
Warranties don't bother us. At the end of the day if something dies in the server it is still going to go down and cost money regardless. The cost isn't in the piece of kit that has died. The cost is in the downtime of that server and the affect on customers that depend on it. A fancy HP/Dell warranty isn't going to improve that. We would rather the responsibility be ours than some faceless organisation. That way we can order in the replacement part(s) on next day shipping or hell run to the nearest computer store to buy them straight away. We keep a couple spare PSUs just in case. And all hard disks are running at least RAID1 or better. Everything is UPSed and air conditioned so I really think the possibility of a motherboard/CPU/RAM etc failure to be very minimal anyway.
I think sysadmins have a bit of a stigma around this area. They like shiny Dell/HP servers as it is less hassle for them and it's not their neck on the line if something they specified in the server goes wrong. It's a bit like saying you own a BMW 7 series versus a Westfield kit car. There is a very different "image" attached to both. That's fine. But understand that not everyone subscribes to that philosophy and couldn't give a toss about having perfectly symmetrical racks with all identical brand servers in them.
I don't really see the issue. If you know what you are doing and know the quality of hardware you are putting together then go for it. We stuck to all 100% Intel chipset kit just to minimize any niggly driver problems.
Currently we're running single PSUs in them but there's nothing stopping us upgrading the PSUs to some dual redundant ones at a later date.
We came close to going with high-end desktop hardware. But in the end we opted for a decent server/workstation class motherboard (think along lines of Supermicro or Tyan). The key thing is to choose the right motherboard.
Warranties don't bother us. At the end of the day if something dies in the server it is still going to go down and cost money regardless. The cost isn't in the piece of kit that has died. The cost is in the downtime of that server and the affect on customers that depend on it. A fancy HP/Dell warranty isn't going to improve that. We would rather the responsibility be ours than some faceless organisation. That way we can order in the replacement part(s) on next day shipping or hell run to the nearest computer store to buy them straight away. We keep a couple spare PSUs just in case. And all hard disks are running at least RAID1 or better. Everything is UPSed and air conditioned so I really think the possibility of a motherboard/CPU/RAM etc failure to be very minimal anyway.
I think sysadmins have a bit of a stigma around this area. They like shiny Dell/HP servers as it is less hassle for them and it's not their neck on the line if something they specified in the server goes wrong. It's a bit like saying you own a BMW 7 series versus a Westfield kit car. There is a very different "image" attached to both. That's fine. But understand that not everyone subscribes to that philosophy and couldn't give a toss about having perfectly symmetrical racks with all identical brand servers in them.
Associate
- Joined
- 24 Jun 2007
- Posts
- 1,869
- Location
- Landan.
Someone should tell the 99.9999% of companies that use pre-built servers that it's just their sysadmins wanting symmetrical racks that's costing them a bomb.
Using custom built servers, in the vast majority of scenarios (as I've said twice now, it only becomes economical in the 100/1000s of servers market), is a complete waste of time, money and effort. To think otherwise, and to attribute the trend of buying custom built servers to sysadmins aesthetic needs, is extremely naive and ignorant.
Here's just a few things to consider:
- How will you replace your 'custom' motherboard in a years time? In two years time? Will everyone still be stocking them? No, they won't - which will mean you'll have to replace all the major components;
- Pre-built servers are tested for 1000s of hours in all typical configurations;
- All the major manufacturers offer same/next-day on-site support/hardware replacement;
- The huge workforce that specifies, engineers, and support servers at the likes of Dell have vastly more experience than you and your colleagues could ever have;
- "Everything is UPSed and air conditioned so I really think the possibility of a motherboard/CPU/RAM etc failure to be very minimal anyway" - I'm afraid this statement highlights your naivety. Sure, your two servers may provide faultless operation for the entirety of their serviceable lifetime. But you make five or ten of them, and I'll bet a fair deal of money you won't be so lucky.
I'll stop there - chances are that you won't concede so it's a futile thing. Needless to say, everyone who works in enterprise IT will disagree with you
Using custom built servers, in the vast majority of scenarios (as I've said twice now, it only becomes economical in the 100/1000s of servers market), is a complete waste of time, money and effort. To think otherwise, and to attribute the trend of buying custom built servers to sysadmins aesthetic needs, is extremely naive and ignorant.
Here's just a few things to consider:
- How will you replace your 'custom' motherboard in a years time? In two years time? Will everyone still be stocking them? No, they won't - which will mean you'll have to replace all the major components;
- Pre-built servers are tested for 1000s of hours in all typical configurations;
- All the major manufacturers offer same/next-day on-site support/hardware replacement;
- The huge workforce that specifies, engineers, and support servers at the likes of Dell have vastly more experience than you and your colleagues could ever have;
- "Everything is UPSed and air conditioned so I really think the possibility of a motherboard/CPU/RAM etc failure to be very minimal anyway" - I'm afraid this statement highlights your naivety. Sure, your two servers may provide faultless operation for the entirety of their serviceable lifetime. But you make five or ten of them, and I'll bet a fair deal of money you won't be so lucky.
I'll stop there - chances are that you won't concede so it's a futile thing. Needless to say, everyone who works in enterprise IT will disagree with you
Done both and within reason there isn't any real life differences. Have backup procedures in place, check hardware to see if any issues are likely. However, as pointed out above, a Dell/HP/whoever server will give you much less hassle if there is a problem.
Someone should tell the 99.9999% of companies that use pre-built servers that it's just their sysadmins wanting symmetrical racks that's costing them a bomb.
Using custom built servers, in the vast majority of scenarios (as I've said twice now, it only becomes economical in the 100/1000s of servers market), is a complete waste of time, money and effort. To think otherwise, and to attribute the trend of buying custom built servers to sysadmins aesthetic needs, is extremely naive and ignorant.
Here's just a few things to consider:
- How will you replace your 'custom' motherboard in a years time? In two years time? Will everyone still be stocking them? No, they won't - which will mean you'll have to replace all the major components;
- Pre-built servers are tested for 1000s of hours in all typical configurations;
- All the major manufacturers offer same/next-day on-site support/hardware replacement;
- The huge workforce that specifies, engineers, and support servers at the likes of Dell have vastly more experience than you and your colleagues could ever have;
- "Everything is UPSed and air conditioned so I really think the possibility of a motherboard/CPU/RAM etc failure to be very minimal anyway" - I'm afraid this statement highlights your naivety. Sure, your two servers may provide faultless operation for the entirety of their serviceable lifetime. But you make five or ten of them, and I'll bet a fair deal of money you won't be so lucky.
I'll stop there - chances are that you won't concede so it's a futile thing. Needless to say, everyone who works in enterprise IT will disagree with you
The cost of replacing the server entirely is much lower when it is a VM host. Pretty much a case of 1. building the server 2. installing Hyper-V 3. some minor config and copying over the VM guests if necessary 4. hey presto back in business. As opposed to with baremetal servers and having to reinstall and rebuild the precise configuration of whatever software was on there. That is a big difference.
If we could have afforded to buy all singing all dancing Dell/HP kit, we would have. But we couldn't. So what do you, in all your infinite wisdom, propose that we do? Just let our business hit a brick wall and stop expanding? Get real.
It's not a custom motherboard. A Dell/HP contains a custom motherboard. We built ours from standard off-the-high-shelf kit. Worse come to worst, if a motherboard dies in 4 years time. We replace the server with a brand new one. Then buy a replacement motherboard off an auction site and have the broken server up and running again within a month. I don't really see the issue?
You need to work for a small business to understand that you can't just blow £15k in a single month on shiny Dell/HP kit and have the absolute most perfect setup. We can just about afford to spend £1k every other month on server kit. So that is what we are doing. See: Cash flow
Associate
- Joined
- 24 Jun 2007
- Posts
- 1,869
- Location
- Landan.
This part, I have no problem with - if you can't afford the specification you require with Dell/HP kit then fair enough - as you say, to simply do without is not an option
And with a two server deployment, you could well be fortunate. My point is that it's wrong to suggest that custom servers are the way to go, it's a false economy for the vast majority of cases. If you're a small business that truly requires systems with 24GB of RAM and huge processing power, then I'd say you were in the very exclusive minority. I don't know of many (or rather any at all) 'traditional' small businesses/start-ups that require such high specced servers. What is it that you do?
And I do/have worked with/in/around small businesses all my life.
Custom servers are a calculated risk and a useful option to have for a small business that happens to have one or more employees experienced enough to make it work.
The risk factor involved, in my opinion, is significantly reduced if the custom servers are merely used as VM hosts.
No business, not even a small business, should IMO have just one VM host. Always have a hot spare. Especially if they're custom servers.
We are more a software house than anything server related. We have a software product that we sell using a SaaS model. Each customer instance of the product can use between 50-200MB of RAM on average. So the more customers we have, the more RAM we need to keep on top of things. The CPU power isn't much of an issue. It is spikey, not perma-load like say a render farm or number cruncher app.
The risk factor involved, in my opinion, is significantly reduced if the custom servers are merely used as VM hosts.
No business, not even a small business, should IMO have just one VM host. Always have a hot spare. Especially if they're custom servers.
We are more a software house than anything server related. We have a software product that we sell using a SaaS model. Each customer instance of the product can use between 50-200MB of RAM on average. So the more customers we have, the more RAM we need to keep on top of things. The CPU power isn't much of an issue. It is spikey, not perma-load like say a render farm or number cruncher app.