SAN and RAID 5

[Shivkumar]

Hi,

Is SAN superior than RAID 5 and other traditional mirroring technologies ?

How traditional LUN different from the LUNs used in SAN ?

Thanks,
Shiv

[Tim Nelson]

The word SAN that you are using is a very generic term for Storage Area Network. A network of storage now a days typically via fibre channel network and one or more disk arrays sold by dozens of vendors of varying performance and funtionality. In general most "arrays" come with some type of memory caching ability and underlying disk and disk controllers.

Your term of RAID 5 is another general term in the sense that it is a method of providing fault tolerence for a disk failure. This method is typically provided in most vendors arrays as well as being supported with JBOD using raid controller HBAs in the host server itself.


You may wish to describe in detail what you are really looking for to get better inputs on these generic terms.

[Shivkumar]

We are working on improving I/O performance of our build and release which is java and C++ based applications.

Some folks are suggesting that creating different LUNs will improve performance. Some people are suggesting some RAID technologies.

I am thinking that SAN will be better option but do have the technical data to support my points.

Appreciate if someone can list superiority of SAN than traditional older technologies. Ours is a big company and cost do not matter to serve our business needs.

Thanks,
Shiv

[Torsten.]

The details may vary from one array to another, but **in general** a RAID0+1 is always faster than a RAID5. RAID1 one will write the same data to 2 disks (in parallel, more or less) but RAID 5 have to calculate the parity information.
As said before, details may vary, but in general ...

[Bill Hassell · Accepted solution]

I would not waste time trying to find a disk fix for your code. If you code reads and/or writes data inefficiently, recommending a faster disk is not going to make happy customers. Whether the application is a database or custom code, you have to profile your code to see where the inefficiencies are located. This is especially true for Java which is notorious for massive CPU usage without proper tuning. Once the code has been adjusted, look at reducing I/O counts with internal caches and wasteful re-reads of the same data.

Here is a great set of Java resources:

http://www.hp.com/products1/unix/java/?jumpid=go/java

Look at HPjmeter and HPjtune.