Disk performance vs Disk size 2

[Guest_imported]

I am currently building an FC4500 Array and I have a conflict between my software supplier and the advice I get from my hardware supplier regarding EMC storage.
Basically the software supplier wants to data spread across as many disks as possible and is requesting the use of 18GB disks. The hardware supplier suggests using 73GB disks saying that the arguments for spreading data across spindles are redundant in todays world.
Using 73GB disks will significantly reduce costs as the number of FC4X00 DAEs can be reduced.

Who is right and how can I prove to the software supplier that 73GB disks will be performant?

 

[bjscott]

they are probably both right - the only way you can find out if it works for your application is to try it ..

 

[bjverzal]

Is your application Oracle by chance ? Generally, a 74GB drive is faster than a 18GB drive because of the media density.

Oracle *really* likes to have users spend $$ on disk drives.

BV

 

[rups21]

Hi,
If you go for 72 GB it will definately reduce cost but using 18 GB disk will increase ur performance.Let us assume both disk have 15krpm.
If u use 72GB disk then no of simultaneous access will be less as compared to that of the 18 GB disks.
If you need excelent performance and both the disks are of 15krpm or whatever speed but same, then go for 18 GB disks.
Regards.

Rupesh

 

[d3funct]

Can you get the 74Gb disks and have EMC do a binfile change to break the METAS into 18GB hypers? An infinite number of monkey typing at an infinite number of keyboards will eventually populate the internet.

 

[danamburgey]

It's actually a combitation of truths, based upon not only disk size, but disk speed...

I'm assuming your 18GB drives are 10000RPM, and the 73GB drives are 7200RPM... (cuz that's more than likely)

(d3funct's comment is mute, cuz this isn't a Symmetrix, it's a CLARiiON

Now most DBA's are only familar with local attached disks, therefore they are aware that each invidual disk spindle is capable of X amount of MB/s throughtput. Now that does kinda apply to the FC Enteprise Storage Arrays, but instead of 16-128MB of Cache Ram you would have on a local Raid controller, these Arrays have 512MB-2GB of Cache ram for all host I/O... Which means on Writes your going to be working at RAM speeds, not DISK speeds... and the array has enough CPU power on the back end, to flush the cache to the multiple disk groups. So litterally unless your writing 512MB of changed data EVERY second, there's a good chance your write performance isn't limited by the disks themselves....

Long story short, go with the 73's... more space is good
performace will be less than noticable.