• Fristo (unregistered)

    Post!

  • Bogolese (unregistered)

    Heartwarming.

  • WTF (unregistered)

    TRWTF is that it all worked out suspiciously well

  • TheCPUWizard (unregistered)

    a) Instead of RAID 1 plus RAID 0 being called RAID 10; I recommend that RAID 5 times RAID 2 be used. And just for clarity, I am not talking about: www.imdb.com/title/tt2265171/

    b) RAID 5 writes will almost certainly be significantly slower for Software RAID controllers; however there are a number of high end controllers where the difference is completely negligible due to these calculations (on the order of 0.03% to 0.1%

  • Hannes (unregistered)

    "David then forwarded his story on to us at The Daily WTF as a reminder that sometimes things actually do turn out okay in the vast WTF-land that is the IT industry."

    I call shenanigans. This story clearly is just made up! FAKE NEWS!

  • WTF (nodebb)

    Things actually turn out okay?? You missed April 1st by a couple of days guys.

  • Quite (unregistered)

    From Kieran's visible attitude with William, one wonders whether David's observations are the last of a whole series of straws breaking the back of the management camel.

  • akozakie (unregistered) in reply to Quite

    Especially since introducing the new position almost seems like a test - as if someone above had enough warnings and decided to hire someone experienced from the outside and see what that person will say after a few days. Nice solution - if everything was ok, the manager wouldn't even notice the distrust, his team was actually strengthened!

  • Fact Checker (unregistered)

    I call BS on this. There's no way a PHB like this would allow his underlying (the sysadmin) contradict and make him loose face in front of a new recruit. Also where's the HTML meta commentary that gives more back story.

  • Commander Koloth (unregistered)

    Succession by assassination. I approve, although the story was clearly missing the mandatory bat'leth duel to the death.

  • Paul (unregistered) in reply to TheCPUWizard

    "however there are a number of high end controllers where the difference is completely negligible due to these calculations" It's not the parity calculation (a 10 year old Intel CPU can XOR gigabytes per second), it's the read-modify-write-cycle. You have to read all the blocks in a stripe, modify the one bit that changed in the database, calculate a bit of parity and write two blocks to disk (the one with the parity and the one where the changed bit resides).

    There are ways to (partially) mitigate that (copy on write and buffering writes (at least the unsynchronized ones) for example) but hardware raid usually doesn't do that.

    Depending on the memory available to the database server (and limits set to the database software) the performance of the APPS-share might be acceptable, provided there's enough RAM to cache most of it, but I still wouldn't do that for a company that can afford a SAN...

  • Kashim (unregistered)

    Makes me wonder what kind of "brief investigation" happened. Outside group of tech people brought in to look at the existing policies? Someone outside trying to FTP into the "firewalled" server? Or maybe they just talked to all of the developers, most of which probably knew that things were bad and hadn't said anything...

  • Publius (unregistered) in reply to Paul

    What isn't clear to me is why RAID 10 would be faster for I/O. If you have 6 disks in RAID10, you are striping across 3 disk. In RAID5, you are striping across 5. Wouldn't reading/writing 60% more disks in parallel make the XOR calculation a drop in a bucket?

  • Dom (unregistered) in reply to Publius

    If you're doing sequential IO that is true, and RAID5 will give you more throughput for that workload, since you are writing entire stripes at a time, and the only overhead is the XOR, which is trivial. With random IO, you are are generally only modifying one unit in the stripe for each request, so you have to read either (a) all the other data disks in the stripe or (b) the old value of this data strip and the current parity in order to determine what the new parity value should be. So your overhead in service time is that you need to perform at least two reads before your modify-write, versus just writing to two drives, so it in absolute terms takes more than twice as long, and your overhead in aggregate is that you are performing at least double the number of operations for each requested IOP - which is guaranteed to be lower than the RAID10 one, and probably much lower, if you are using the simpler read-whole stripe technique, which I believe is more common.

  • BKay (unregistered) in reply to Publius

    True but it entirely depends on how much data you want to write. If you had a stripe size of 128KB and a RAID Level 5 with 5+1 disks, everytime you write at least 640KB in consecution, you don't need to read the parity stripe, because it will be overwritten anyway. RAID5 will be faster because you can write to 5 disks at the same time. Now, if you had a large amount of data to write, the dedicated parity disk will become the bottleneck (because for each 640KB block you have to write one 128KB stripe to this particular disk). However, the cache of the RAID controller will probably get this. Vice versa, if you only have 5KB to write, you'd read a stripe from each of the six disks in order to calculate parity and write back to six disks.

    Which makes RAID5 good for file servers or other applications handling big amounts of data, but not so good for database servers where only bits change all over the storage area.

    Personal story: We had one server go down once because of a defective RAID(5). I discovered that the controller was configured incorrectly and thus did not warn of the drive failure. Upon repairing I found that nearly all disks had a huge amount of bad sectors. There was the exact same server, same age, same disks, same configuration - and when I checked it, all disks had ZERO sector problems. One was hosting a SQL database and the other was a file server.

  • BKay (unregistered) in reply to BKay

    Addendum: Not to be misunderstood - the dedicated parity drive is obsolete in any modern controller, it was called RAID4 then. I just mentioned it as an example when large data could be a bottleneck. Also, controllers nowadays don't read all disks to calculate the parity information, so the performance impact isn't as huge as it once was.

  • William (unregistered)

    I swear I ordered RAID 10!

  • Dobby the IT Elf (unregistered)

    Am I the only one who thought this was going in a completely different direction? I could have sworn the twist would be that the PHB ordered two RAID 5s thinking that would make RAID 10...

  • wtf (unregistered)

    TRWTF is David randomly running disk benchmarks on a production server. He should be fired immediately.

  • partial defense (unregistered)

    In his partial defense, the FileZilla installer was known to package adware/spyware a few years ago. It might have been the fault of where FileZilla was hosted at the time.

  • Joe (unregistered)

    What about the Linus RAID failure where they striping 3 in software?? RAID 5 hardware cards

  • Zenith (unregistered)

    TRWTF is everybody believing that William was the one who got fired.

  • Ashley Sheridan (unregistered) in reply to Fact Checker

    @Fact Checker why on earth would the PHB loosen his face? I mean, I get that older PHBs can have wrinkles, but surely loosening a whole face would be just too garish for the typical working environment?

  • Ali Razeghi (google) in reply to wtf

    Fired for running a very quick disk benchmark in that situation? William, is that you again?

    You do realize the disks themselves probably hit max load often anyways and running a very quick benchmark to note the problem is nothing compared to the long term performance impact, right? William you were rightfully fired in this situation, even though the guy you put through hell had to run a benchmark to get it through.

  • Pista (unregistered)

    I see that thedailywtf.com needs a raiding on the spammers. I suggest hiring William, he seems to know how to raid. And he can build firewalls, too!

  • Offf (unregistered)

    TRWTF is David not being fired as being too competent. This is how real management works.

  • Ugh (unregistered)

    You had me until you mentioned SAN. That they didn't immediately move to AWS RDS or some other managed DB and instead had to deal with LUNS and figuring out which DB files need to be on which spindles, etc. means they were wasting a lot of time doing administration they didn't have to do. And God-forbid that Hitachi SAN has a hiccup and your Oracle DB goes in the shitter.

  • Herr Otto Flick (unregistered) in reply to wtf
    TRWTF is David randomly running disk benchmarks on a production server. He should be fired immediately.

    You are obviously unaware that every OS can tell you how many transactions per second each disk is doing...

    $ iostat Linux 2.6.32-573.26.1.el6.x86_64 (dev-xxxxxxx.xxxx.xxxx) 31/03/17 x86_64 (6 CPU)

    avg-cpu: %user %nice %system %iowait %steal %idle 4.40 26.79 6.44 0.03 0.81 61.52

    Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn vda 4.65 15.04 245.41 297937086 4861662986 vdb 31.16 50.05 818.03 991571900 16205305942

  • Ethenrosethomaz (unregistered)

    one of the best download bigo live online chatting application

  • Mile (unregistered)

    This is one of the best articles as I have found here, I think you should have create more New stuff here Newintrends.com

  • Jyoti (unregistered)
  • Saniya (unregistered)

    It's very helping post and I would like to say that your writing is so unique and useful for us.Thanks for sharing with us . escorts in Jaipur

Leave a comment on “Raiding the New Manager”

Log In or post as a guest

Replying to comment #:

« Return to Article