Forum Stats

  • 3,838,603 Users
  • 2,262,384 Discussions
  • 7,900,692 Comments

Discussions

Size and number of luns for the database (500GB or 1TB)

2»

Answers

  • aitorit0
    aitorit0 Member Posts: 52 Blue Ribbon
    edited Jun 1, 2018 8:10AM

    Omg... another variable for define the size of disks, the AU size... 4 disks of 3TB for data diskgroup seems not good idea for our DB

    Following this note: Lun Size And Performance Impact With Asm (Doc ID 373242.1)

    "For larger LUNs it is recommended using a large ALLOCATION UNIT."

    But we have a OLTP db with random access workloads (not datawarehouse), and following this thread:Master Note for Automatic Storage Management (ASM).

    "For OLTP configurations, 4MB is an optimal value (as used by Exadata by default)"

    Then,

    Do you recommend for OLTP small luns?

    Which is the formula?

    Which is the minimum and maximum size for OLTP?

    thanks

  • Levi Pereira
    Levi Pereira Member Posts: 2,676 Gold Trophy
    edited Jun 1, 2018 10:16AM

    Think about Lun performance and management. Not only about performance.

    Because create Large or Small Luns will have minimal impact about performance but huge impact about management.

    aitorit0
  • Dude!
    Dude! Member Posts: 22,830 Black Diamond
    edited Jun 1, 2018 9:59AM

    There is no one fits all, or best solution. Larger AU sizes typically provide performance advantages for data warehouse applications that use large sequential reads. However, it will decrease performance if you don't. So it all depends on your database and how you use it - there's no performance formula.

    AU allocation  isn't fixed and will automatically increase based on demand. The more extents the database allocates, the larger the AU size will grow. I suggest  you review the Allocation Units section below:

    https://docs.oracle.com/en/database/oracle/oracle-database/12.2/ostmg/asm-intro.html#GUID-BC612D35-5399-4A35-843E-CF76E3…

    For most installations the default AU size of 1 MB should be fine. If it wasn't, the default would be larger.

    Why do you think you have to tune ASM for better performance? What's wrong with the defaults?

  • aitorit0
    aitorit0 Member Posts: 52 Blue Ribbon
    edited Jun 5, 2018 7:38AM

    Ok, let me imagine a crisis situation.

    Imagine that we have arrived to 98% of total capacity, and we need to add a lun urgently to diskgroup.

    Yes, I know that is a very bad situation and we need to know why is not warned us before.... but we need space now...

    What is faster, add a lun of 500MB or add a lun of 1TB to the diskgroup?

  • EdStevens
    EdStevens Member Posts: 28,778 Gold Crown
    edited Jun 5, 2018 7:43AM
    aitorit0 wrote:Ok, let me imagine a crisis situation.Imagine that we have arrived to 98% of total capacity, and we need to add a lun urgently to diskgroup.Yes, I know that is a very bad situation and we need to know why is not warned us before.... but we need space now...
    What is faster, add a lun of 500MB or add a lun of 1TB to the diskgroup?

    Neither.

    Why do you think the size of the LUN is a factor in adding it to a disk group?

  • aitorit0
    aitorit0 Member Posts: 52 Blue Ribbon
    edited Jun 5, 2018 1:06PM Answer ✓

    what I would like to say is that is faster add a "small disk" (in a crisis time) because the data to rebalance is less, let me explain with an image:

    pastedImage_1.png

    Ok, I know that if you need to add another 512GB you will take the double of time, but I only think in the rebalance time operation and the affectation to the production database.

  • Dude!
    Dude! Member Posts: 22,830 Black Diamond
    edited Jun 5, 2018 2:51PM

    That's OK from the perspective of used space, but you have to put into consideration that copying 200 GB from 8 devices can be slower than copying 400 GB from 4 devices. You probably won't notice a difference between 4 or 8 devices, but at a certain point the number of devices and channels can add more performance overhead than benefit, in particular when there are shared or logical drives or partitions that actually reside on the same physical device.

    What sort of performance gain do you expect? How would anybody know how many devices your storage system can efficiently manage? There are simply too many unknown variables. You're only looking at the number of devices and disk space, but there is a lot more involved than these two aspects.

    If you feel more comfortable with 8 x 512 GB than 4 x 1 TB, by all means. I doubt it won't make any noticeable performance difference, but is more likely going to increasing your maintenance cost, double the space, energy consumption and risk of failure.

    aitorit0
  • EdStevens
    EdStevens Member Posts: 28,778 Gold Crown
    edited Jun 5, 2018 3:05PM
    aitorit0 wrote:what I would like to say is that is faster add a "small disk" (in a crisis time) because the data to rebalance is less, let me explain with an image:

    While I agree with Dude! in that your calculations are over-simplified, I would point out that your efforts are better spent making sure you don't get into a "crisis time" in the first place.  One of the very first things I did when I took my current postition was to address this very issue of finding myself in 'crisis time' over lack of disk space.  Each of my servers has a daily job that collects disk usage stats and writes to a log file.  I then have an Excel spreadsheet with a macro that pulls those log files from the various servers, parses them out, and loads the data into various individual worksheets, where they are then charted and trendlined.   I keep a year's worth of history It only takes me a minute each day to run the data collection macro and assess where we are and where we are headed.  If I see an unsual spike, I address it.  If a trendline shows me running out of disk within the next 2 years, I start alerting management so that they can factor the additional requirement into the next budget cycle.

    While it is good to be prepared for a crises, it is better to pro-actively remove the possibility of said crises.

    aitorit0
This discussion has been closed.