back to article Screwed by the cloud: Hardware vendors looking for that raison d'refresh

Four in ten of the servers currently residing in datacenters across the globe are at least six years old. Meanwhile, not only does this 40 percent slice consume 66 percent of the energy used by all bit barns – they also only provide 7 percent of the world's total compute. This is according to Uptime Institute research outlined …

  1. Doctor Syntax Silver badge

    Weasels strike again

    "Four in ten of the servers currently residing in datacenters across the globe are at least six years old. Meanwhile, not only does this 40 percent slice consume 66 percent of the energy used by all bit barns – they also only provide 7 percent of the world's total compute."

    They always think we can't see what they're up to. In this case we're supposed to think that 66% of DCs' energy is producing 7% of the DCs'output. But the world's computing power is not all in DCs so it's a false comparison.

    As a matter of interest hat %age of the world's and DCs' computing power and DC energy is mining cryptocurrency and training LLMs?

    1. jdiebdhidbsusbvwbsidnsoskebid Silver badge

      Re: Weasels strike again

      Exactly what I was thinking. If this 40% of servers only delivers 7% of the world's compute, perhaps it would be best to ditch them and do without them at all. I'm willing to give up 7% of my personal compute budget to save a load of carbon emissions. Anyone else?

  2. Anonymous Coward
    Anonymous Coward

    Bean Counters

    He needs to direct this to the bean counters. Though they'd just fob it off and capex vs opex vs budget lines and so on.

    If they can extend an assets life they are happy and if it is doing its intended job, they don't care.

    Certainly at some companies I've worked at, a good spec server on will get repurposed a few times and each time into less taxing roles to really extend the assets usefulness.

  3. Pascal Monett Silver badge

    So, HPE has found another excuse for a raft of upgrades

    Anything is good to make customers spend their money, eh McDonald ?

    Who paid you for this report ? Do you really think I believe that you're being objective ?

  4. Paul Crawford Silver badge

    One ProLiant Get 11 server can replace eight ProLiant Generation 8 machines, he said, giving the same compute but 19 percent less power consumption. This is before improved security and manageability are brought into the mix, McDonald pointed out.

    Only 19% less power? So the Gen11 is consuming about the same as about 6.5 of the Gen 8 servers?

    Also no word on what is insecure about the older servers, more shitty iLOM stuff?

    In terms of cost I guess you need to factor in the worker time related to physically swap and re-image the machines, but presumably data centre operators have that all down to a polished solution by now.

    1. doublelayer Silver badge

      I don't think they're suggesting that the newer machine is more secure by design, but that if you're running eight servers, then theoretically that's eight opportunities to mess up the security of one of them, so you should definitely be trying to have as few servers as possible, and wouldn't you know we happen to have one server that can be like a bunch of older servers in one box. Of course, if you are running a lot of different servers, then you're very likely deploying the same image to most or all of them, meaning you have exactly as many opportunities to do it wrong. At that level, you also probably deploy a bunch of VMs on that server and weird manual configuration is going to be on specific VMs, meaning the number and size of physical servers is irrelevant to security at that level. Customizing each server manually is more likely to happen if you only have a few of them anyway, in which case buying an expensive newer model may be overkill because you aren't using as much performance as you just bought.

    2. Tom Womack

      Saving 19% of electricity is a spectacularly trivial bit of opex to try to use to justify an enormous capex - saying that you have to rent an eighth as much expensive DC space would seem more plausible.

      A big chunky new dual-Granite-Rapids server might use as much as 2kW; 19% of that is 400W, which is less than £2000 a year even if your industrial electricity contract came from Dewey, Cheatem and Howe; but the server costs at least ten times that and HPE are explicitly saying that you can't expect it to last the decade it would take for the lowered opex to repay the capex.

  5. the spectacularly refined chap Silver badge

    What if these 40% of servers are also providing 90% of the storage? Little point upgrading them if they can still fill their network links. Sure replace the disks, the servers themselves, that's less important.

    An exclusive focus on compute suggests deliberate tunnel vision.

    1. Tom Womack

      If you have any machines left with twelve 15krpm SAS drives in, you probably are saving money and electricity immediately if you replace them with a couple of SAS eight-terabyte SSDs. But since they stopped making 15krpm SAS drives in 2016, if you're in that situation you almost by definition don't care about your storage estate.

  6. JamesTGrant Bronze badge

    I rather suspect that ‘old infrastructure’ is fulfilling dedicated workloads which, on average, consume a low(ish) percentage of the peak potential CPU throughput (two/three reasons they’d still be alive after 6 years!)

    If you care about power consumption then you can allow low power modes/states even in really old HPE servers - the Watts per computed output overhead scales pretty linearly with computed output and provided you don’t need the throughput to be capable of instantly jumping, you can save a load of heat and money.

  7. Pelican Express

    Simply add RAM to old servers?

    We run everything on the cloud. Our workloads run in Kubernetes clusters. Metrics show that the most constrained resources is memory (generally 75 to 85% usage) while CPU usage % is lower around 30%

    It is more cost effective for us to double memory rather than CPU. We still choose more recent CPU generation because it just happen that their cost is same or lower than older generation.

    If the older CPU cost was lower we would definitely go for older CPU.

    1. Tom Womack

      Re: Simply add RAM to old servers?

      Second-hand DDR4 RAM is incredibly cheap nowadays - I have 288GB in a home-lab machine, ECC DDR4 is down to £1.50/GB second-hand even in 64GB sticks.

      Ah, but you're in the cloud so you don't get to make that tradeoff, and the fact that the most recent CPU generation is really quite expensive per CPU is a matter for Amazon's fearsome purchasing department rather than for you.

  8. This post has been deleted by its author

  9. Sora2566 Silver badge

    Lies, damn lies, and statistics

    Can't help but notice that they said "66% of the power of *bit barns'* power but 7% of the *world's* compute".

    How much of the *bit barns'* compute do they provide, then? Because I'm willing to bet it's a lot higher than 7%.

  10. harrys Bronze badge

    lies damn lies and statistics ... from a sales guy just doing his job, fair enuf :)

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like