Re: Question on disk contention

Melvin Davidson <melvin6925@xxxxxxxxx> · Thu, 31 May 2018 10:09:16 -0400

On Thu, May 31, 2018 at 10:04 AM, Ron <ronljohnsonjr@xxxxxxxxx> wrote:

    On 05/31/2018 08:52 AM, Melvin Davidson wrote:

          On Thu, May 31, 2018 at 1:13 AM,
            Charles Clavadetscher <clavadetscher@xxxxxxxxxxxx>
            wrote:

            Hi
              Melvin

              As an answer to a previous post you wrote:

              "Also, your main problem is that when you have two exact
              same queries executing at the same time, they will cause
              contention in

              the disk, and neither one will make much progress."

              Could you elaborate a little more on the meaning of
              "contention in the disk"?

              What is it that happens?

              Thank you and have a good day.

              Regards

              Charles

          >Could you elaborate a little more on the meaning of
          "contention in the disk"?

          >What is it that happens?

        To simplify, you have two users/jobs,
          both wanting the exact same information. So the system
          instructs the disk to get 

        that information from the disk, which
          causes the disk head to "seek" to the position of the first
          eligible row and 

        continues positioning to other eligible
          rows. Now the job is not exclusive, so the system temporarily
          switches to the
        other job, which causes the disk to go
          back to the first row and work from there. The switching back
          and forth continues,
        so that instead of one job finishing
          quickly, they both have to take turns waiting for needed
          information. That takes
        a lot longer,

        Try this, Select a table that has a lot
          of rows, ideally 1M+. Then start a query with a WHERE clause
          and see how long
        it takes. Then submit the same query
          from 5 separate connections simultaneously and see how long
          that takes.

    Why isn't the OS caching the disk blocks, and why isn't Postgres
    using the cached data?

    -- 

      Angular momentum makes the world go 'round.

>

    Why isn't the OS caching the disk blocks, and why isn't Postgres
    using the cached data?

It does, but the cache is for each connection/job. They are not shared.
-- 
Melvin Davidson
Maj. Database & Exploration
Specialist
Universe Exploration Command – UXC
Employment by invitation only!