Home / server / Questions / 129485
Accepted
Berkay
Asked: 2010-04-06 11:09:32 +0800 CST

How to define a layered model for Storage Architecture?

  • 772

i don't have enough knowledge about storage systems but have to tell my audiences about how storage works in organization data centers.

For this purpose to keep things simple, first i want to show them a layer from top to bottom and then i want to explore all these layers.

What i mean by layered model in storage architecture?

i want to start my presentation something like that, first layer can be applications, application request a data from storage and then ... ... (the topics i want to cover are file sytems, metadata,the physical implementation of storage (das, nas etc.) ... the request comes to the storage device and storage device (depends on the technology) goes through the disk and disk send back the data using iscsi protol, we can say that the iscsi protocol managing the path between... ...

if you help me explain these steps layer by layer by examples and names of some key technologies to be defined, really appreciate it.

thanks.

storage datacenter

2 Answers

  1. Best Answer

    You're in luck, because there's a GREAT blog called SysAdmin1138 Expounds that just covered this.

    The series of posts is called "Know your I/O".

    Intro

    The Components

    The Technology

    Caching

    I really recommend you check them out. I'm certain that it'll help you and your audience.

    • 5

  2. Matt already pointed to it, but the link titled "The Components" shows the full stack from bottom (disk drive) to top (file-based application) in more detail than you probably care about. It's all about the abstraction. The article on Caching goes into some detail about when writes are reported as committed to applications, since that can vary due to a number of reasons. The article on Technology goes into some details about SATA vs SAS drives, which I found interesting to write.

    What I didn't go into much detail on are filesystem specific details such as metadata handling, journaling, or large-directory handling. That depends on each filesystem individually, and undergoes constant change (in the linux world anyway). Generally speaking, journaled filesystems write metadata to the journal first, and then commit actual data; which means that slowdowns in metadata processing can slow down writes even if your storage I/O channels are clean. For an example of how this can work on NTFS, I go into it here.

    • 2