3

We're currently in the process of adding additional servers to our website. We have a pretty simple topology planned: A Firewall/Router Server infront of a Web Application Server and Database Server.

Here's a simple (and technically incorrect) diagram that I used in a previous question to illustrate what I mean:

Diagram

We're now wondering about the specs of our two new machines (the Web App and Firewall servers) and whether we can get away with buying a couple of old servers. (Note: Both machines will be running Windows Server 2008 R2.)

We're not too concerned about our Firewall/Router server as we're pretty sure it won't be taxed too heavily, but we are interested in our Web App server. I realise that answering this type of question is really difficult without a ton of specifics on users, bandwidth, concurrent sessions, etc, etc., so I just want to focus on the general wisdom on buying old versus new.

I had originally specced a new Dell PowerEdge R300 (1U Rack) for our company. In short, because we're going to be caching as much data as possible, I focussed on Processor Speed and Memory:

  • Quad-Core Intel Xeon X3323 2.5Ghz (2x3M Cache) 1333Mhz FSB
  • 16GB DDR2 667Mhz

But when I was looking for a cheap second-hand machine for our Firewall/Router, I came across several machines that made our engineer ask a very reasonable question: If we stuck a boat load of RAM in this thing, wouldn't it do for the Web App Server and save us a ton of money in the process?

For example, what about a second-hand machine with the following specs:

  • 2x Dual-Core AMD Opteron 2218 2.6Ghz (2MB Cache) 1000Mhz HT
  • 16GB DDR2 667Mhz

Would it really be comparable with the more expensive (new) server above?

Our engineer postulated that the reason companies upgrade their servers to newer processors is often because they want to reduce their power costs, and that a 2.6Ghz processor was still a 2.6Ghz processor, no matter when it was made.

Benchmarks on various sites don't really support this theory, but I was wondering what server admin thought.

Thanks for any advice.

Community
  • 1
Django Reinhardt
  • 2,286
  • 3
  • 38
  • 58
  • If you buy 2nd hand, you don't get any kind of support. I'm assuming this is for a business need, so uptime and support might be important... Depends if you give a toss about making money. Don't be cheap. Just buy new hardware. Every time. – Tom O'Connor Dec 16 '10 at 15:08
  • One question is, if you down-spec a new server to about the same actual performance as this supposedly cheap second-hand machine - where does the price land? My guess is, close to the second-hand price... though it's still a guess based on my specific locale's second hand market ^^ – Oskar Duveborn Feb 20 '11 at 15:56

5 Answers5

11

First off, a 2.6GHz processor is not a 2.6GHz processor if they're from different generations. You're correct in thinking twice about that. This has been true for a long time now (at least since the 486 / Pentium days), and so it's important to point out to your engineer just how wrong the Megahertz Myth is. Especially given the massive performance improvements i7 based chips offer over Core / Core2 based ones at the same clock speed.

That being said, that's not my first concern with this plan. My first concern is that used servers will have a significantly reduced operational life that a new server, since you don't know how it was previously used, under what conditions, or what'll happen to it in transit on the way to you. Generally speaking, for production systems, reliability should always take precedence to performance, since it'll cost you way more to fix a dead production server than to upgrade a server that's too slow.

My feeling is that the price difference would have to be very, very substantial to even want to look at doing this, and that if you're buying used, you'll want to redundantly cluster them just to be safe.

Graeme
  • 343
  • 1
  • 6
  • As a web developer, this is the only answer here I'd upvote. Reliability is king. Have a longer operational life and good (or preferably great) warranty support is typically way more important to me than initial hardware cost or performance. (Run the numbers: How much does being down 1 hour cost you? 2 hours? 4 hours? 12 hours? At what point does downtime outcost the price difference? What's your acceptable risk? Etc.) – John Rudy Nov 06 '09 at 20:04
  • That's precisely what I thought. It's a shame there isn't an indepth article explaining this (that Wikipedia article is very weak). Thanks a lot for reminding me what's important: Reliability. – Django Reinhardt Nov 06 '09 at 20:12
  • I've always heard that if a server does not fail within the first few months of operation, it is a lot less likely to fail at all within it's service life. In my experience, the vast majority of servers will still be functional when they are finally decommissioned, typically meaning five or more years in service. – Roy Nov 06 '09 at 21:11
  • Roy: generally this is true, under normal operating conditions. However, I've been down the used server road before, and generally the servers that get sold used (rather than just repurposed for other tasks) are the ones that've been run in "bad" environments: no cooling, flaky power, etc. These servers generally can and will fail prematurely, and the problem is, you don't know when buying used if that's what you're getting. Considering how cheap servers are now (under $2k for an i7 system with 12GB of RAM and 6TB of RAID10), it's not worth the risk, IMO. – Graeme Nov 06 '09 at 21:21
  • Django: Take a look at this article on Ars detailing the architectural differences between i7 and Core2 processors. It should help explain on currently shipping hardware how newer design allows a CPU to run faster at the same clock speed. http://arstechnica.com/hardware/news/2008/04/what-you-need-to-know-about-nehalem.ars. – Graeme Nov 06 '09 at 21:33
  • @Roy: While generally this may be true, the reality is I've personally seen well-maintained kit give up the ghost late into its warranty period. Typically what I've seen are PSUs and mobos fry. Admittedly, it's rare, but again: Is an assumed-business-critical app worth the risk? It's one thing if it's Aunt Nelly's recipe database; it's quite another if it's sales goal/planning data for a Fortune 500! :) – John Rudy Nov 06 '09 at 21:36
  • 1
    As a rule of thumb I've found that used hardware is never worth the trouble and worry it's going to cost. At the very least you'll have to double up to ensure availability (which usually puts you right back up to the cost of new) – Jim B Nov 06 '09 at 22:14
  • @Jim B: I agree in principle, though I'd say that used hardware as a spares pool, rather than functioning in a production capacity, can be a really cheap hedge against unexpected failure. A 4-hour 24x7 support agreement is nice, but walking into the storage room, grabbing a spare part from an eBay box, and bringing the production box back up while the replacement part is on the way is even better. – Evan Anderson Nov 06 '09 at 23:57
2

Buy 2 of the AMD's ( the price still be under the newer one ) make them redundant so if one fails the other will take the load, even new servs fail ( don't even get me started on IBM X3650 )... You can even use one of them with ESXi or whatever you prefer ( i'm XEN fan ) and you will have a VM on the second one to play with that will have a lot of free resource while the other operates fine. You will also have same parts for both and even if both fail sim. ( very unlikely ) you will have a chance to fix it without waiting for ordered parts or having to search for parts. "Reliability is king"-redundancy is the army behind the king ;)

set ekh
  • 129
  • 1
  • 4
1

Well I think that it at first depend on "HOW MUCH MONEY YOU HAVE TO INVEST?" For example I'm setting up a on-line game and I unfortunately don't have enough money to get new servers Rack 2U that cost 4.500 euro each.

I need 2 heavy processing machines, one for WEB and the other for DB and I need another to make backups.

If I need to get all those 3 machines I estimated that I'm going to spend about 10000 Euro but I don't have that.

So I opted instead of getting new servers to get used and get not just 3 machines but 5 for ONLY 2.500 Euro.

In this way I got 2 WEB servers, 1 DB, 1 Backup, and one for spare in case something goes wrong.

I think that if you have a company that can invest you have to get new stuff but second hand servers that are GREAT could make possibile the realization of projects for those who can't afford the heavy costs but have the knowledge to start the job.

When you enter in IT without much money you need to understand and know what you really need because yes new is always better but if you know what you need you can get great machines for really low prices that can build up an empire.

Instead of spending 10K euro I saved money, got more servers even spare one and payed Internet connection for the first year and payed my developers.

Hope this help

Nixphoe
  • 4,584
  • 7
  • 34
  • 52
Mauro De Vitti
  • 11
  • 1
1

There's likely not a huge difference in performance between the two, so if that's your only concern it might be worth it to try for the second-hand AMD box. Still this depends greatly on the application.

Based on experience from my AMD vs Intel benchmarks for Oracle OLTP databases, I think a dual Opteron 2218 is even likely to outperform a single X3323 for memory or I/O intensive applications.

Roy
  • 4,376
  • 4
  • 36
  • 53
0

Sounds like a reasonable thing to do. Web servers them selves don't need tons of horse power. A dual core machine with 16 gigs of ram sounds sufficent. If this were a database server, I would go the other way though. Databases you know need lots of processing power to grind through the data, and more processors/more ram on those machines is necessary.

Good luck in your purchasing and hope this helped you.

Chris
  • 800
  • 1
  • 7
  • 10