New (to me) HP Proliant Microserver G7 N54L Build... RAM upgrade, boot device, additional drive bay (non hot-swap)..

  • Heavens! Don't forget about SOLID data backup. With drive failures being inevitable, It would be a shame to do the work to consolidate that much data and risk losing it.

    I have sets backup HDDs stored at my office and at my parents. At one point I had considered LTO tape but HDD prices kept dropping. Insane that 6TB can be had for $150 these days.


    Sent from my LM-G710 using Tapatalk

    • Offizieller Beitrag

    Wrong. At least if you prefer buying cheap and don't rely on ECC RAM. The more DRAM you have the more often bit flips will happen. If you like (silent) data corruption this is a great recipe for just that.
    https://www.zdnet.com/article/…nightmare-on-dimm-street/

    One can have both - ECC and low cost.


    For some of the older tech (which a G7 N54L is) used 4GB ECC UDIMM sticks can be had for around $20 to $25. After a few days of testing with memtest86, if they work they work.

    • Offizieller Beitrag

    I have sets backup HDDs stored at my office and at my parents. At one point I had considered LTO tape but HDD prices kept dropping. Insane that 6TB can be had for $150 these days.
    Sent from my LM-G710 using Tapatalk

    You're definitely set, even with off-site backup. :thumbup:


    Years ago I looked at tape backup but they tend to be slow and, when compared to just using another hard drive, not worth the hassle.

    • Offizieller Beitrag

    I'm not gonna take sides here but I have to agree with @tkaiser the beast I have is the old school server 4GB Ram single Xeon processor and it served 250 + users and was running Windows Server 2008.


    The new one has 32GB Ram and an 8 core Xeon running Windows Server 2012 with 6 VM's one of those VM's requires a minimum of 8GB (which turned into 12GB minimum in the end, runs a 'heavy' db and finance software)

    While it's off topic; in a campus environment, I wouldn't be OK with 250 users working into a single physical server. When I was involved in this sort of thing (you know back when smoke signals were used), I worked in environments with seriously "demanding" users. Since the budget wasn't an issue, with that many users, I'd want a cluster or at least more physical servers. (Outages due to a single point of failure, for that many users, would have been out of the question.)


    That was then - this is now. Currently, I have no idea what good design would be for such a topology :huh: , but I imagine that cost is a serious constraint at a school.
    ____________________________________


    On the RAM question, I'm running two boxes with ZFS. One has 4GB and the other has 12GB. As it seems, if RAM is available, a good sized chunk goes into the page cache. Interestingly, the 12GB box has less free ram on average, than the 4GB box, but I have a client backup server running on the 12GB box.


    4GB (OMV4)


    12 GB (OMV3)

  • For some of the older tech (which a G7 N54L is) used 4GB ECC UDIMM sticks can be had for around $20 to $25. After a few days of testing with memtest86, if they work they work.


    Nope. Again: zdnet.com/article/dram-error-rates-nightmare-on-dimm-street/


    These are reported errors from ECC DRAM in the field. If you do not have ECC DRAM then of course also bit flips happen from time to time. But there is no mechanism that tells you 'DIMM 3 threw 2 errors within the last 24 hours'. With the primitive ECC DRAM we use today single bit errors are corrected (so causing no harm) and you're notified of single and double bit errors if you look in EDAC logs or better setup a monitoring for stuff like this. Only with 2 bit errors that are very unlikely on normal modules the result is a bit flip in memory.


    Without ECC RAM these single bit errors also happen all the time but always result in bit flips. The result of a bit flip can be nothing at all (if you were lucky) or data corruption, a program crash, filesystem corruption or even a kernel panic or a BSOD when running Windows.


    New servers here run 72 hours memtest, then we fire them up and integrate them into the monitoring system. Correctable 1 bit errors do occur from time to time. Without ECC DRAM you won't even know. And usually symptoms look like software issues. That's most probably also reason N°1 so many users think they could save some bucks and choose non-ECC RAM.

  • I have a N54L with 4x3TB and a 4TB drive in the DVD bay. I use it for SnapRAID mainly.


    I don't recall having to use any custom BIOS to boot from the DVD bay drive, just change BIOS boot order.


    I used gparted to make a small partition on that drive for the OS - you can run the OS from the internal USB port on a thumb drive but not worth risking as all the re-writing will eventually kill it. The rest of the 4TB drive is for SnapRAID parity.


    I got ECC ram for reasons tkaiser stated. I put a small USB3 card inside but I've never got it running reliably to be honest, I gave up trying to sort it as I rarely need to import/export via USB anyway, so USB2 will do.

    OMV 6 on Raspberry Pi 3/4/5, Odroid C2/HC2 and three cobbled together x64 boxes running Snapraid.

  • you can run the OS from the internal USB port on a thumb drive but not worth risking as all the re-writing will eventually kill it

    That's what the flashmemory plug in is for. Together with disabled monitoring almost no writes happen any more.


    And as usual: we're dealing here not with something magic but with a simple Linux system where we can look at what's going on. If you do an 'apt install sysstat' and then fire up an 'iostat' you'll see how much data has been read and written from/to each partition. It's easy to spot what's happening.


    The only downsides when dealing with low-end flash products like USB pendrives are

    • you have no clue how Write Amplification looks like (and that's the most important thing)
    • you have no SMART support to query the drive about wear out indication
    • you can't use TRIM which is really bad once the amount of data exceeded the drive's native capacity

    That said we run all small servers from either USB sticks or better SD cards (there you can TRIM at least from time to time using "SD Formatter" app).

    • Offizieller Beitrag

    I know you love to do this (really I do) - :) - but what's with the "Nope", and how would that apply to the "quoted" post?


    As was stated, I was referring to ECC UDIMMS. I've bought used 4GB ECC UDIMMS for about $20 to $25, thoroughly tested and they work fine. I've queried them and found that ECC has, indeed, corrected a bit flip here and there. So, I'll say it again; if using older tech - both "ECC" RAM and "low cost" can be had.

  • how would that apply to the "quoted" post?

    'if they work they work' -- ECC RAM even if tested successfully needs monitoring since otherwise it's the same as with RAID without monitoring. Once a disk dies or a module starts to get bad the whole redudancy is of no use any more since as soon as 2 bit memory errors occur the same might happen as without ECC RAM: data corruption, a program crash, filesystem corruption or even a kernel panic.


    If you use ECC RAM just to get a good feeling (same reason the majority of OMV users plays RAID) then you don't need to care. If you care about your data you need an eye on syslog or EDAC logs.

    • Offizieller Beitrag

    And yet, the accuracy of the original post remains -

    One can have both - ECC and low cost.
    For some of the older tech (which a G7 N54L is) used 4GB ECC UDIMM sticks can be had for around $20 to $25. After a few days of testing with memtest86, if they work they work.

    This is just extraneous spin on it ----|
    ............................................v

    'if they work they work' -- ECC RAM even if tested successfully needs monitoring since otherwise it's the same as with RAID without monitoring. Once a disk dies or a module starts to get bad the whole redudancy is of no use any more since as soon as 2 bit memory errors occur the same might happen as without ECC RAM: data corruption, a program crash, filesystem corruption or even a kernel panic.
    If you use ECC RAM just to get a good feeling (same reason the majority of OMV users plays RAID) then you don't need to care. If you care about your data you need an eye on syslog or EDAC logs.

    Have you got any more ideas from Data-Centers, for home users and small businesses? :) If it's that important, how about documenting it in a "How-To" or an implementation guide, versus dropping hints in a thread? Publish it.
    ________________________________________________


    Along a similar line, one could say; managed network devices, maybe with an HP Openview console, are not just for the data center,,, imagine what could be done with them at home.. If one really cares about network throughout, congestion, and fault monitoring...


    Nope,, heavy vlan segmentation, multi-layer routing/switching, and the monitoring thereof, may have a place in a data center, but it's not needed or even appropriate at home.

Jetzt mitmachen!

Sie haben noch kein Benutzerkonto auf unserer Seite? Registrieren Sie sich kostenlos und nehmen Sie an unserer Community teil!