Server refresh - when should I?
-
So, the server infrastructure at our main office is three HP DL380 G6s, running ESXi 5.1 with local storage only, plus a 4TB ReadyNAS 2100 for extra storage for the file server.
Two of the DL380s were purchased in July 2010 and one was purchased in June 2009. The NAS was purchased in 2011.It's all working fine. We haven't had any major software upgrades or business growth, so everything is working as well as it was when first installed, although there's is less free disk space than I'm truly comfortable with. Several faulty disks have been replaced plus a RAID controller battery, but we've had zero unplanned downtime. Carepacks are around $1k per server per annum.
I was just wondering when the best time to replace is. Is 5 years too old for a server?
I'm definitely replacing the oldest server within the next 12 months, as it only has 146GB drives and 12GB RAM (it only runs one VM which is our ERP system), is over five years old now, and I need some extra storage and RAM to install our new ERP system and run it in parallel with our old ERP system.
I had assumed we would all be cloud based by now, but we're not there yet. At the same time, I'm reluctant to invest too heavily in our physical on-premise infrastructure as I don't know how many more years such a physical environment will be valid. Our new ERP system will be Microsoft Dynamics NAV and I was hoping to go for a hosted solution, but hosted appears to be considerably more expensive than on-premise at the moment, so I'm not sure it will happen.
Is there any sound reason for replacing 5 year old servers? I'm not a hardware guy, so haven't a clue. How long do Proliants last before they start failing too regularly? At the moment, they all seem to be working fine, and I'm inclined to think if it ain't broke don't fix it.
-
Below link should give you an answer
http://creativecomputerms.com/2013/10/why-should-i-replace-my-out-of-warranty-server/Especially this one: The life-cycle of a server should be no greater than the manufacturer’s warranty
On a side note: We had server which lasted for almost 8 years before anything broke, but when it broke, parts were not available, so we had to wait for quite some time to get it back up and running, luckily, it was not a critical server, so we survived!
-
That link doesn't really help at all I'm afraid, as HP provide post-warranty maintenance agreements, so the servers won't run out of warranty any time soon. I don't know how long HP continue to support Proliant servers, but it's a lot longer than the initial 3 year warranty that comes with them. I also don't know how long VMWare will continue to support older servers (anyone know?). I certainly wouldn't expect to replace new servers running ESXi or other hypervisors after just 3 years. One of the advantages of virtualisation is the ability to safely extend hardware life.
-
Five years is most definitely not too old for a server. In the enterprise space eight years is common and ten years isn't unheard of. At some point you pay too much in maintenance and that is really what makes the swing as long as everything else is still good.
-
@ambarishrh said:
Below link should give you an answer
http://creativecomputerms.com/2013/10/why-should-i-replace-my-out-of-warranty-server/Especially this one: The life-cycle of a server should be no greater than the manufacturer’s warranty
On a side note: We had server which lasted for almost 8 years before anything broke, but when it broke, parts were not available, so we had to wait for quite some time to get it back up and running, luckily, it was not a critical server, so we survived!
Where those HP, IBM or Oracle servers though? Getting parts after ten years on any if those is easy. They will give you decade warranties.
-
A lot if the feeling that servers shouldn't run for 8+ years comes from the SMB use of low end gear especially in 1990-2003 on low end Intel IA32 procs. That was cheap stuff and really did not last.
Beyond that the move from IA32 to AMD64 architecture made equipment obsolete long before it began to fail. And a few years later virtualization extensions did the same thing. But those were anomalies. Normally servers last 8-12 years pretty easily. Eight or nine for AMD64 platforms, ten plus for Power, Sparc and Itanium.
-
I think this is definitely a much more difficult call today that it was 5+ years ago. Inside the last 5 years we've seen a much lower need for more processing power than we had available to us when the original purchased was made (i.e. we bought more compute power than we needed).
The idea of keeping a server running as long as you have warranty is a valid one, but as CB mentioned, you can generally buy extended warranties anywhere from 5-7+ years total. Server hardware is definitely manufactured better today in the x86 SMB segment.
I'd look at the following factures:
Do you need more compute power?
Do you need more disk space? Can you move to a NAS or perhaps storage expansion units hanging directly from a server (not SAN)
Can you consolidate?
What is your RTO? example - if you lost a machine, could you restore the VMs to another host you have, and live with the performance until a new host was purchased?Consolidation might be one of your greatest advantages here. Sounds like you could easily get away with two if not even one modern server. You'll probably need to go with tiered storage for performance reasons otherwise a ton of smaller disks.
-
Good questions. I do need more space and memory, which is the main reason I was going to replace the oldest server. However, I'm now thinking I can just replace the eight 146GB drives with eight new 300GB drives and upgrade the RAM from 12GB to 32GB. That wouldn't cost too much and should give me everything I need.
I can, and have, restored VMs to another host. I like having 3 servers, rather than one, for redundancy. Two servers might be the ideal, but then you have to run them at a maximum of 50% capacity to allow for redundancy, whereas with 3 you're running them at 66% capacity. I'm not sure about the advantages of consolidation, but need to consider it some more. I'm not comfortable about mixing older G6 Proliant servers with a new Gen8 server - I prefer all our servers to be the same generation, but I'm not sure if it makes any difference.
If you're confident the servers will last for 8 years, then that's what I'll do! I should have moved to Office 365 by then, at least. Thanks for the advice as always.
-
Upgrading the RAM and the drives is probably 60-70% of the cost of a complete server replacement. The last purchase I made, the server and second power supply was under $3K compared to the nearly $4k for RAM and disk.
Consolidation will probably bring lower power bills, require less cooling (i.e. lower power bills) and fewer parts to maintain. In your situation I'd never personally want to go to less than two hosts.
Considering the advances in compute power, a single host can probably handle your entire environment right now.
-
@Dashrender said:
Upgrading the RAM and the drives is probably 60-70% of the cost of a complete server replacement. The last purchase I made, the server and second power supply was under $3K compared to the nearly $4k for RAM and disk.
I'll need to do some sums. But given that we can capitalise a new server over 5 years, but couldn't capitalise new RAM and disks for the old server, a new server may make the most sense financially.
-
There's plenty of things to factor in. How energy efficient are the new servers? If it's a healthy gap, you're losing money by keeping the old stuff running, and once you do the required math, you'll have a better idea on the lost cost savings by waiting. It might be a little, or it might be a lot, depending on the types of servers and configuration. Also, there's the consideration of density. A G8 has much more computing power than a G6, allowing it to hold a greater number of VMs. Theoretically, you could reduce the number of servers total, requiring less management overhead and operational cost.
Why are you still on ESXi 5.1? Your vSphere environment should be under maintenance. If your hosts are on the HCL for 5.5, go ahead and upgrade ESXi. If the hosts aren't on the HCL, it's a good sign that it's time to upgrade.
-
New servers typically make sense over upgrades. Upgrades have a huge price premium on parts, new servers have discounts.
New servers are smaller, faster, more reliable and more power efficient.
-
@alexntg said:
Why are you still on ESXi 5.1?
Because upgrading requires coming in on a Sunday and doing an upgrade that I'm not completely comfortable with. Basically, I don't like working Sundays and always put it off. It's not like I'd get paid or get time off in lieu, or if anyone I work with would even know I'd worked on Sunday.
TL;DR...I'm too lazy.
-
@scottalanmiller said:
New servers typically make sense over upgrades. Upgrades have a huge price premium on parts, new servers have discounts.
New servers are smaller, faster, more reliable and more power efficient.
Really, is there much of a difference between the 6th and 8th generation Proliants? They seem to have similar parts to me. I can see that moving to solid state drives will make a difference (but that's still very expensive), but otherwise?
-
@scottalanmiller said:
Five years is most definitely not too old for a server. In the enterprise space eight years is common and ten years isn't unheard of. At some point you pay too much in maintenance and that is really what makes the swing as long as everything else is still good.
I second his comment. Scott talked me into keeping our PowerEdge 2900 as a second virtual host and I am going to use it for development and application serving. We had some doubts about it not being up to the task but I turned on virtualization and it is working great. Though my RAID card says the battery needs replacement.
-
@Carnival-Boy said:
@scottalanmiller said:
New servers typically make sense over upgrades. Upgrades have a huge price premium on parts, new servers have discounts.
New servers are smaller, faster, more reliable and more power efficient.
Really, is there much of a difference between the 6th and 8th generation Proliants? They seem to have similar parts to me. I can see that moving to solid state drives will make a difference (but that's still very expensive), but otherwise?
There are improvements, yes. Cheaper, faster parts mostly. More cores at faster speeds for less money. And the ILO4 is sweet. It's not a huge leap. But old memory is super expensive.
-
Any issues migrating a VM from a G8 server to a G6 (and vice versa). I recall there can be issues when running different CPUs, although am I right in think that this only effects vMotion? We don't have vMotion, and so shut down the VMs and start them up on the new host.
-
@Carnival-Boy said:
Any issues migrating a VM from a G8 server to a G6 (and vice versa). I recall there can be issues when running different CPUs, although am I right in think that this only effects vMotion? We don't have vMotion, and so shut down the VMs and start them up on the new host.
If it's a cold move, you're fine. vMotion is where it gets tricky, and there's even ways around that.
-
@Carnival-Boy said:
Any issues migrating a VM from a G8 server to a G6 (and vice versa). I recall there can be issues when running different CPUs, although am I right in think that this only effects vMotion? We don't have vMotion, and so shut down the VMs and start them up on the new host.
As long as you are migrating cold you can move from an HP Intel box to a SuperMicro AMD box and it won't care. Only hot migrations are an issue.
