Consolidated Business & Enterprise Computing Rant Thread

Discussion in 'Business & Enterprise Computing' started by elvis, Jul 1, 2008.

  1. GumbyNoTalent

    GumbyNoTalent Member

    Joined:
    Jan 8, 2003
    Messages:
    9,495
    Location:
    Briz Vegas
    Kenisis is the issue, US-EAST-1 Region.
    https://status.aws.amazon.com/
    Would have been funnier if it was next week during re:Invent (AWS Conference).
     
  2. PabloEscobar

    PabloEscobar Member

    Joined:
    Jan 28, 2008
    Messages:
    14,410
    Compute is being redirected to handle black friday/cyber monday deals.

    You just can't afford it right now.
     
    elvis and GumbyNoTalent like this.
  3. tobes

    tobes Member

    Joined:
    Dec 23, 2001
    Messages:
    4,029
    Location:
    Melbourne
    Yup, so much stuff runs through US-EAST it is crazy.

    But there are 5 availability zones so what can possibly go wrong.
     
  4. wintermute000

    wintermute000 Member

    Joined:
    Jan 23, 2011
    Messages:
    2,520
    Doesn't IAM and/or Azure AD both run out of a single region under the covers?
     
  5. GumbyNoTalent

    GumbyNoTalent Member

    Joined:
    Jan 8, 2003
    Messages:
    9,495
    Location:
    Briz Vegas
    Early Black Friday processing, that makes sense, COVID-19 lockdowns and Black Friday early marketing... BOOM! more load then AWS Kinesis can handle in that Region.
     
  6. caspian

    caspian Member

    Joined:
    Mar 11, 2002
    Messages:
    11,536
    Location:
    Melbourne
    my next door neighbour is a systems dev for KMart. he's looking a bit frazzled at the moment.
     
  7. cvidler

    cvidler Member

    Joined:
    Jun 29, 2001
    Messages:
    14,620
    Location:
    Canberra
    Microsoft hasn't admitted it, but it's been reverse engineered by some smart people, that AzureAD (world wide) is entirely reliant on the Texas region/AZ.
     
  8. kjparker

    kjparker Member

    Joined:
    Jun 28, 2001
    Messages:
    1,471
    Location:
    Sydney
    that's not concerning at all is it...
     
  9. millsy

    millsy Member

    Joined:
    Mar 31, 2007
    Messages:
    13,381
    Location:
    Brisbane
    Got more information on that, sounds like an interesting read!
     
  10. NSanity

    NSanity Member

    Joined:
    Mar 11, 2002
    Messages:
    18,244
    Location:
    Canberra
    When will the cloud for my cloud be up so my cloud can be up?
     
  11. power

    power Member

    Joined:
    Apr 20, 2002
    Messages:
    65,569
    Location:
    brisbane
    no clouds, only rain today.
     
    Hive likes this.
  12. millsy

    millsy Member

    Joined:
    Mar 31, 2007
    Messages:
    13,381
    Location:
    Brisbane
    They need to make sure to implement RAICP* for their outage pages

    *RAICP - Redundant Array of Inexpensive Cloud Providers
     
    Hive likes this.
  13. 2SHY

    2SHY Member

    Joined:
    Aug 10, 2010
    Messages:
    7,756
    Location:
    Sydney NSW Australia
    I hear IBM / Oracle have some cloud space available if you are desperate for something.
     
  14. wintermute000

    wintermute000 Member

    Joined:
    Jan 23, 2011
    Messages:
    2,520
    [​IMG]
     
    GumbyNoTalent likes this.
  15. NSanity

    NSanity Member

    Joined:
    Mar 11, 2002
    Messages:
    18,244
    Location:
    Canberra
    Woah now buddy. Inexpensive? Bitch that I stands for Independent.

    MultiCloud(tm) - Share that bread.
     
  16. millsy

    millsy Member

    Joined:
    Mar 31, 2007
    Messages:
    13,381
    Location:
    Brisbane
    I agonised over a suitably pointed word for I, and independent never came to mind. Son I am disappoint (in myself)
     
  17. NSanity

    NSanity Member

    Joined:
    Mar 11, 2002
    Messages:
    18,244
    Location:
    Canberra
    Apparently inexpensive vs independent is heavily debated. For 20+ years i've known it as Inexpensive, about 5 weeks ago during internal training, I was made aware of the alternate.

    If a big chunk of your company's business is selling best of breed enterprise storage, they don't want people associating your product with inexpensive if its not that - who knew?
     
    millsy likes this.
  18. cvidler

    cvidler Member

    Joined:
    Jun 29, 2001
    Messages:
    14,620
    Location:
    Canberra
  19. OP
    OP
    elvis

    elvis Old school old fool

    Joined:
    Jun 27, 2001
    Messages:
    43,461
    Location:
    Brisbane
    Really good conference on today from the Pawsey Centre's Computational Fluid Dynamics team (lots of guests from Oil and Gas and other large fluid dynamics mobs) on what they call (and pardon the marketing-heavy term) "interactive supercomputing". These very problems are being worked on both with hardware and software vendors.

    Some talk about this stuff which is "old" now:
    https://www.mellanox.com/products/GPUDirect-RDMA

    But way more talk about newer tech that aims to get the data to the GPU much, much faster than before. They all note that their billing systems were the first to tell them how idle their GPUs were most of the time, as they hung around waiting for data to arrive. But there's new stuff coming (AMD are on board, as are Nvidia), to make that problem less of an issue.

    https://en.wikichip.org/wiki/amd/infinity_fabric
    https://www.anandtech.com/show/1559...hitecture-connecting-everything-to-everything
    https://rocmdocs.amd.com/en/latest/index.html + OpenMP on GPU

    The desire is always real time, which nobody expects, but it's a valid goal to strive for. But at any rate, the consensus is GPUs are too fast and too valuable to ignore, and solving the IO problem at all levels is more worthwhile than just pushing orders of magnitude slower CPUs to go 20% faster.

    I stick by my prediction that workloads will move to more purpose-specific compute, and the era of the general purpose CPU being the be-all and end-all of large scale computing will vanish. Not overnight of course, but I would bet my career that the list of folks you've mentioned in the quoted post won't be buying the "same thing but 20% faster" on their next major refresh. The architecture will look very different.

    [edit]

    Some anecdotal talk too of what Space-X are doing to better use GPUs.
     
    Last edited: Nov 26, 2020
  20. NSanity

    NSanity Member

    Joined:
    Mar 11, 2002
    Messages:
    18,244
    Location:
    Canberra
    who'd of thought that the company that nvidia owns, doing cool stuff would have Nvidia onboard.

    Jim Keller still thinks this stuff is edge case. Tesla did something different for their "not-self-driving-but-really-actually-just-fairly-good-assisted-driving" solution, because of commercial costs (we're gonna build a zillion cars, so lets own the stack - like they do so many components).

    I'll see if i can dig up his talk/presso - but it basically boiled down to the fact that a lot of purpose-built asic's have a usable lifespan of a few years at best, as the problem moves on. But to meet performance/energy targets, the cost to actually make some silicon is ramping exponentially against a dwindling number of fab's.

    Hence with the market for CPU's (and to a lesser extent, GPU's) - it makes sense to throw development bucks there. Also - when that problem no longer needs to be solved/can't be accelerated - you can at least do something on that > $100m Supercomputer. Rather than a pile of useless ASIC's that don't solve any problems today.
     
    Last edited: Nov 26, 2020

Share This Page

Advertisement: