Stefan Radtke

...welcome

  • Increase font size
  • Default font size
  • Decrease font size
Home XIV Storage Blog
XIV Storage Blog

Increasing Parallel IO on Oracle and DB2

E-mail Print

Although it's mentioned often, the workload for an XIV Storage System should produce much more parallel IO than on traditional systems to leverage XIV's strenght. In many of my projects, customers are using DB2 or Oracle as their prefered database and some of their historically grown applications do not use the mechanisms that their DBMS provides to increase the throughput and minimize response time of their queries. I am (unfortunately) not a database expert but I would like to point to some easy techniques that Oracle and DB2 provide to maximize parallelism. Most of these techniques don't require heavy changes in the application.

Read more...
 

Improving Oracle RMAN performance on XIV

E-mail Print

I had recently a customer who wondered why his RMAN backup jobs ran slower on the the XIV than of his FC based strorage system. The expectation was that is should run much faster since the XIV in (R 2.2) can write more than 1000MB/s. As mentioned several times, XIV is a parallel storage system and the more parallel IOs, the better the performance. This is also true for RMAN.

Here is how we increased the RMAN backup performance up to the desired throughput:

Read more...
 

Dynamic Storage WebCast

E-mail Print

WebCast vom 27.4.2010 zum Thema Dnamic Storage – Wie intelligente Speicherstrategien die Organisation entlasten und die Prozesse beschleunigen!
Den gesamten WebCast kann man nach kurzer Registrierung hier ansehen. Im Webcast hatte ich die Gelegenheit die XIV-Technologie entwas näher zu erläutern.

Hier eine Mini-Vorschau zum Webcast:

 

 

Ankündigung zum XIV-Webcast

E-mail Print

Am 27.4.2010 um 10:30 werden wir einen Webcast zum Thema Dynamic Storage – Wie intelligente Speicherstrategien die
Organisation entlasten und die Prozesse beschleunigen
senden. Dort werden wir natürlich aus das Thema XIV diskutieren.

Die Teilnehmner können während des Webcasts mit uns chatten und Fragen stellen, die wir während der Sendung beantworten.

Bitte registrieren Sie sich hier zum Webcast: http://event.on24.com/r.htm?e=205362&s=1&k=AB2B2B803BD778FFF3110403080305AB

 

Coming up: XIV Webcast

We' ll launch a free webcast about Storage Technology with a focus on XIV.

The webcast will be sent on Tuesday, 27.April.2010 at 10:30 - 11:30.

Visitors of the webcast can also chat with us and ask questions.

Please register for the webcast at: http://event.on24.com/r.htm?e=205362&s=1&k=AB2B2B803BD778FFF3110403080305AB

The Webcast will be in German language.

 

Impact on number of IO threads

E-mail Print

In the processor and server technology we currently see a move towards multi-threaded architectures and applications. However, (unlike XIV) the traditional storage systems currently do not reflect this. Since it's parallel architecture, XIV supports multi-threaded and parallel workloads much better than traditional storage systems. As said before, the more IO you can bring to XIV, the better the performance. That can be done through connecting a lot of servers (virtual or physical ones) or use applications with multi-threaded IO. As you can see in the following graph, XIV's IO-throughput increases significantly with the number of IO-threads on the applications site.

Read more...
 

OLTP Performance

E-mail Print

Doing an OLTP benchmark requires an appropriate environment and setup which I have rarely access to. Especially with the parallel XIV architecture we need several servers to be able to drive XIV to its limits. However, I often get questions for people running SAP or similar applications who want to know how much IOPS XIV can handle. As an idication we can use the results from the other test where I showed the impact on IO threads to XIV. This test has been run with a random 70/30/50 workload of 8k Blocks. The following graph shows the results.

Read more...
 

General XIV Performance Considerations

E-mail Print

With the tradional dual-controller storage systems, the storage media has -as we all know- massive impact on the storage system performance. This is because all IOs have to be handled by one or two controllers and their single caches. However, the impact of the media is less important with a parallel storage system like XIV since it does IO over all the modules and it has a distributed cache where all caches are local. No cache syncronization is requried. Therefore, XIV can deliver High End performance with SATA disks. There are some things to consider which I would like to repeat here:

Read more...
 

XIV Performance

E-mail Print

In this section I will share some expierences regarding XIV-performance from the field. Don't expect extended benchmark results here but you may find usefull information, hints and tricks and probably indications of whether XIV meets your requirements.

I'll try to cover the following topics over time (I'll do that during my free time so don't expect to see updates every day):

These are my ideas for now relating XIV performance. There might be other intersting things over time which are worth discussing them here. If you would like to follow the blog you may subscribe to my whole XIV blog or just the performance section with your RSS-Reader.

Please remember that all I write here are my personal thoughts, oppinions, and experiences not official IBM statements.

 

Binary and decimal Capacity conversion

E-mail Print

I get sometimes asked why the volumes appear smaller in the OS than in the XIV GUI or CLI. The reason is simple: ...

Read more...
 

Reclaiming Space in Thin Provisioning

E-mail Print

A lot has been said and written about Thin Provisioning [1],[2],[3]. Thin provisioning in XIV allows overprovisioning of storage pools. That is, a pool can have a hard size of 1TB and a soft size of 2 TB, allowing you to create volumes worth of 2 TB capacity (although physically we have only 1TB).

If one uses Thin Provisioning there will be the desire to "reclaim" space if files have been deleted. That is usually a problem since if one removes a file, only the inode table (UNIX) or the allocation table (Windows) is modified. The data blocks itself still remain in the filesystem the storage. The filesystem knows that

Read more...
 

Another Reliability Test: Module Failure

E-mail Print

In my previous blog about the reliabilty test of a customer, we modified the test because the rebuild was extremely fast. We were somewhat sceptical about the good result so we increased the amount of

Read more...
 

Warning: Illegal string offset 'active' in /homepages/34/d229784961/htdocs/radtke/templates/ja_purity/html/pagination.php on line 129

Warning: Illegal string offset 'active' in /homepages/34/d229784961/htdocs/radtke/templates/ja_purity/html/pagination.php on line 135

Warning: Illegal string offset 'active' in /homepages/34/d229784961/htdocs/radtke/templates/ja_purity/html/pagination.php on line 129

Warning: Illegal string offset 'active' in /homepages/34/d229784961/htdocs/radtke/templates/ja_purity/html/pagination.php on line 135
  • «
  •  Start 
  •  Prev 
  •  1 
  •  2 
  •  Next 
  •  End 
  • »


Page 1 of 2