back to article Real talk: Why are you hanging on to that non-performant disk?

Generations of change have produced layers of storage that are a challenge to manage. When I was a boy, storage was easy. You had servers with internal hard drives that had capacities of tens of megabytes, and that was it. It was inefficient – unused space on one server couldn’t be used by the other server – and it was …

COMMENTS

This topic is closed for new posts.
  1. sorry, what?
    WTF?

    "If you can afford a new car then you’re not going to buy a beaten up second hand vehicle..."

    That's definitely worded to be emotive ("beaten up second hand") and, when phrased more appropriately, is clearly untrue.

    If I could afford a new car but I can get a decent second hand one (that is only a few iterations older) whilst taking out all that forecourt depreciation I would jump at the second hand one instead. I'm not proud.

    If I had a small business where money needs to be spent wisely, of course I would look to re-purpose existing kit, as long as it makes sense to do so.

    Anyone suggesting you just scrap useful stuff is a) unenvironmental and b) blind to economics. Clearly someone grew up in the "just buy a new one" generation!

    PS: Not that I generally disagree with the article, but I think this metaphor fell short.

    1. Naselus

      Re: "If you can afford a new car then you’re not going to buy a beaten up second hand vehicle..."

      Exactly. The author's point is basically rubbish, since he believes that performance is the only metric which matters for storage. It isn't.

      If I have 4 terabytes of WORN data, then the only metric I care about for it is price per gigabyte. I couldn't care less about it's performance. If I recommended buying a 4TB of flash disks to store it on, then I'd be wasting thousands of pounds of budget that would be better spent elsewhere.

      This isn't even a new discussion. It's long been settled, in fact; only a neophyte stampedes for the fastest possible storage without reference to the workloads.

      1. Peter2 Silver badge

        Re: "If you can afford a new car then you’re not going to buy a beaten up second hand vehicle..."

        "If you can afford a new car then you’re not going to buy a beaten up second hand vehicle..."

        Well, actually...

        I bought a cheap, beaten up near 20 year old second hand vehicle instead of a new one, despite being able to afford to. Why waste money? A visit to a scrapyard yeilded replacements for the "beaten up" bits for less than an hours wages, and it's good as new.

        The yearly cost of maintenance is under the monthly cost of a brand new replacement, and my car is better built, has the same fuel efficiency as something new, and kicks out less than the legal emissions limit all of the time, instead of just when being tested. Most importantly I can spend literially tens of thousands that i've saved on things I do want, instead of a flashy car that II don't particually need. (since my need is to get me to work, and go out places, not to try and show off how much better off I am than the next door neighbors and advertise please rob me in carparks. Nobody breaks into, vandalises or steals 20 year old cars!)

        When it comes to data storage, much the same applies.

        HP Proliant 380 G8 1.6TB SSD, $3,995.95

        HP Proliant 380 G8 2.0TB HDD, $195.95

        Now, $8k for a SSD pair to RAID, or $400 for a pair of larger HDD's resulting in a $7,600 saving that could be applied to buying other equipment. Let me think...

        HDD's are going to be around for at least the next decade precisely because of this. SSD is too expensive to stick everything on, unless your lucky enough to work somewhere where IT is given a blank cheque to spend as much as they want.

  2. TRT Silver badge

    SSD?

    Are you using that to mean "Flash"? Because there are better ways of using solid state storage than SSD. And I think HDD still has a place in on-prem solutions.

  3. Anonymous Coward
    Anonymous Coward

    Really?

    'Imagine phoning your supplier: “Hi, I’d like to buy some really slow storage, please, to augment my nice fast stuff”. “Certainly sir”, they’d reply. “How bad do you want it?”. You wouldn’t buy slow storage, so why compromise performance by hanging on to old non-performant disk?'

    Um... Yes, I would. Not for main storage, of course, but slower storage is frequently used as a backup medium because it is cheaper.

    Even then, if I have a large amount of - say - historical image or document data that is not frequently accessed, then slower, cheaper storage is good.

    Likewise, in some cases Dev and Test servers that use lower data volumes will run fine on slower, cheaper storage. I wouldn't use the high performance shapes for them either. There will always be far more dev/test boxes than Production...

    1. Korev Silver badge

      Re: Really?

      I agree. There are plenty of scenarios where having slow disc is a good compromise between cost and performance. A good example would be a system in a hospital delivering X-Rays, MRIs etc. After the images are taken then the doctors will want to look at the images and will view them from say the last few weeks a lot and the system also needs to be responsive. For older images the medics will still need to access them, but that bit of the storage won't be hammered in the same way.

      1. TRT Silver badge

        Re: Really?

        More so, in a hospital, one might have 12 medical imaging scanners acquiring and dumping scans into a system all at the same time. One needs a system with the bandwidth to cope with all of those systems offloading at once, obviously flash, then it can migrate into slower storage at the system's leisure.

      2. PatientOne

        Re: Really?

        One scenario that came to mind is to replicate a busy storage area, so for testing, as well as general development, slow disk could be very useful.

        1. Naselus

          Re: Really?

          All this, of course, makes it doubly ironic that the author's bio includes the following:

          "His main pastime is shouting at people who try to install technology without considering whether it actually fits the business or the requirement".

  4. Anonymous Coward
    Anonymous Coward

    I'm sure the article called out the new options reducing complexity

    However all these options are themselves complex, even if the complexity is now at a higher level than wrangling storage across individual drive spindles. Moving the problem to an architecture level is not in itself reducing complexity at all, only the length of the grey beard needed to handle the technology is reduced.

    The workload, the costs, and physical environment can influence what is appropriate.

    I have never worked anywhere where a business case for storage replacement hasn't been a great deal of haggling with various PHB's either. This will invariably end up with you can have half of what you want but keep using the older stuff.

  5. Anonymous Coward
    Anonymous Coward

    ...and with protocols such as FCoE

    Seriously?

    At that point you lost credibility.

    1. Naselus

      Re: ...and with protocols such as FCoE

      I think he'd more or less lost credibility when he started a conversation about tiered storage without ever identifying the performance vs capacity dichotomy that sits right at the heart of storage architecture.

      So really, he failed to understand why storage bods exist, and in doing so underlined why they have to.

  6. doublelayer Silver badge

    No? No. No!

    I disagree with pretty much everything here. Allow me to make some points:

    I want SSD storage on everything. It's nice and fast, it seems to crash less, it's small, I can do large-scale stuff without worrying about bottlenecks. It's really expensive. Sometimes I have a cause for large datasets, and I don't have the budget to put tons of stuff on SSD.

    <blockquote>Imagine phoning your supplier: “Hi, I’d like to buy some really slow storage, please, to augment my nice fast stuff”. “Certainly sir”, they’d reply. “How bad do you want it?”.</blockquote>

    Imagine phoning your supplier: “Hi, I’d like to buy some backup tapes, please, to augment my nice fast storage in case someone breaks it”. “Certainly sir”, they’d reply. “We ship those all the time. When do you want it?”. That's really slow storage. Yet, people use it. Why? They use it because it's economical. You don't need to read backups every day unless something's very wrong, so you don't care that it can take a minute.

    <blockquote>You wouldn’t buy slow storage, so why compromise performance by hanging on to old non-performant disk?</blockquote>

    I have slow storage. I don't need to buy it. I probably won't go out and buy a ton of deliberately slow storage, because I have to pay for that. I might as well pay for something that will last longer. But now that I purchased some disks for, let's say, $1000, I have them. I could probably get the same disks for $500 now, but if I need to store some data on the disks, the cost for using the current disks is $0. I hear your article shouting at me that these disks aren't free. Yes, they are. My costs are power (disks versus servers, the power is not a problem), cooling (the same), risk of failure (true, but I keep backups and new disks can fail too), and empty slots in my servers (my disks aren't 74gb ones. I have enough hard drive bays).

    But now you want me to replace cheap or free slow disk with cloud storage. Why? I don't want cloud storage. Maybe if I have to put data into somewhere so that people from lots of places can get it. That's a case for cloud. Maybe it's part of an off-site solution; that's also a case for cloud. But not for speed or cost. I have to pay for cloud storage by the month. If I'm using it to store large data, the kind I'd be putting on slow disk, I have to pay a lot every month. Not to mention that now I have to pay for bandwidth to put it there and bring it back, which is very much not free--even if I am on a connection devoid of ISP annoyance, I'm affecting everyone else by putting so much traffic on the line. I also have to worry about security--if I'm paying the cloud provider, I'm going to make sure they don't inspect my data, but there's always someone else. I encrypt as much as I can, but that's one more headache I can have if I'm using different systems. I assume you also planned on my making backups of this cloud data just in case the cloud provider has a terrible problem and loses it? I wonder what you expected me to use to back the stuff up to? A cabinet full of reserve SSDs that I only use for that? No thank you.

    My stack of old disks is free, because I'm done with the days where people would ask me why I have so much old junk on my desk. I can use them for storage purposes. If they break, I try to get something new to replace them. If I can, an SSD. One of those gigantic samsung ones, please. If that won't work and I need to store a large amount of data, I'll go to my disk cabinet, find an appropriately sized disk, and shove it in. True, it'll be slower, but it allows me to take my finite budget for storage and decide what system deserves the limited number of fast drives I can get, and which systems will have to make do with perfectly functional slow disk. Mending is better than spending--both I and the finance department agree.

  7. myxiplx2

    So many things wrong with this. Sure, we'd all love flash storage everywhere, but when you have a limited budget, or huge requirements it just isn't necessary.

    I've worked on storage refresh projects for schools where the peak requirement is 500 IOPS. They don't need to spend £50k+ on flash storage that delivers 100,000 IOPS. Sure disk is slower but for them, its still fit for purpose.

    Other clients have needed multiple petabytes of storage for archival projects. This data will be read once in a blue moon so the performance requirements are minimal, and the software managing the archive doesn't care if it has to wait for the data. Heck, most of the time it's pulling data from disks that are spun down (30s latency) or tape libraries (tens of minutes of latency). An all flash solution for these customers would be eye wateringly expensive. Quite literally 10's of millions more to purchase, and with far higher maintenance and running costs too.

    The comment about deduplication was also terrible. Whichever VMware bod suggested to run both was basically clueless. The benefit of running two deduplication engines in series like that is generally negligible, but the performance overhead, memory hit, and latency penalty is going to be very apparent.

    If you have centralised storage with dedupe and compression then use that. It will have controllers designed for the purpose. And that allows you to turn off these features in your hypervisors, improving compute efficiency and allowing you to run more VMS per host, with better performance, better density, lower license costs and lower running costs.

    Running dedupe twice is utterly ridiculous.

This topic is closed for new posts.