Storage for DBAs

This page contains an index to an extended blog series I have been writing called Storage for DBAs. As a long-time database guy entering the murky world of storage, I found three were a lot of lessons to learn and myths to debunk.

It’s not a great name for a blog series because it’s not just aimed at DBAs but also architects, developers, administrators of any kind, managers, buyers and anyone with an interested in modern storage. However, in these Twitter-friendly times, the name Storage For Architects, Developers, Administrators of Any Kind, Managers, Buyers and Anyone with an Interest in Modern Storage didn’t seem very catchy.

Index

Introduction

12 Responses to Storage for DBAs

Amardeep Sidhu says:

December 18, 2014 at 11:42 am

Have opened this page many times but could never manage to read the posts. Finally started reading today.

Great stuff.

Thank you !

Reply
Raviteja Chilakamarthi says:

February 25, 2015 at 5:13 pm

Amazing wealth of information.

Reply
Pingback: SQL Performance, Part II — Disk I/O: metrics and scales | Oracle Diagnostician
Aa ab says:

September 1, 2016 at 4:24 pm

Thanks for the articles. You explain an complicated topic with so much simplicity. I learned a lot.

Reply
István says:

September 6, 2016 at 3:41 am

Thank you so much for these articles, they helped me a lot. I just started to digging into data recovery.

Reply
Jane Zhou says:

October 9, 2016 at 7:59 am

The most understandable and best-written technology blog.

Reply
- flashdba says:
  
  October 9, 2016 at 8:04 am
  
  Thank you!!
  
  Reply
  - Jane Zhou says:
    
    October 19, 2016 at 1:54 am
    
    I read through almost all the articles. These are very helpful! I will be a faithful reader continuously.
    
    One message you shared is the power consumption of SSD is lower than HDD. Theoretically yes. Do you have any data point to compare, and whether the interface of SAS, SATA, PCIe will have impact on the power consumption. The reason put forward this is from my experience, 2.5″ SAS HDD usually consumes 11W power, while the latest PCIe SSD power is budgeted at 25W.
    
    Any blog you want to recommend also helps.
    
    Reply
    - flashdba says:
      
      October 19, 2016 at 10:08 am
      
      Hi Jane
      
      The data sheets of SSDs and HDDs will show you a reasonable difference in power consumption, but that’s only half the story. Consider also the fact that many more HDDs are required to meet the same performance levels of a handful of SSDs, particularly for random I/O where IOPS becomes the limitation. Even the fastest HDDs can deliver not much more than 200 IOPS, while a single Samsung PM863 SSD can deliver almost 100,000.
      
      On the subject of comparing specifications, you must always ensure you are comparing like for like. SSDs come in consumer or enterprise-class categories and will have significantly different performance and environmental specifications. For HDDs this is even more the case, with so many combinations and permutations on sale. Be careful you don’t compare the power draw of a mid-range consumer HDD to a high-end SSD.
      
      Finally, I have no insights on the effect of the interface, but what is interesting about SSD power consumption (in my humble opinion) is that the type of NAND flash used (e.g. MLC versus TLC) has a significant effect, as of course does the fact that writes use more power than reads. You can see an interesting graphic displaying this exact phenomenon in this review of the class-leading Samsung PM863 and SM863 solid state drives, which are wildly popular in the enterprise flash array space:
      
      http://www.thessdreview.com/our-reviews/samsung-sm863-pm863-ssd-review-960gb/8/
      
      Reply
Bilal Inamdar says:

November 14, 2016 at 6:31 pm

Hi flashdba..
Is there any tool to measure the io latency, iops, bandwidth to compare two storage system before migration ? I found so many but as you have a great deal of knowledge based on the point of view of oracle dba that i need to ask you before search all over the place.
The tools i found are iostat,iometer,iozone,fio,vdbench,slob,swingbench,benchmarkfactory and many more. I would like to know how do you exactly test for the latency,bandwidth and iops when you do a benchmark. And what exact parameter you use and what you look for in the results.

Regards,

Reply
- flashdba says:
  
  November 14, 2016 at 8:16 pm
  
  You are right, there are many tools. However, it’s important to test the whole platform not just the storage, so if you are building a database platform I would look no further that Kevin Closson’s SLOB.
  
  I do it like this:
  
  https://flashdba.com/database/useful-scripts/using-slob2-for-pio-testing/
  
  Reply
varad says:

April 26, 2021 at 4:01 am

Very Much informative. Thanks Team

Reply