Disk Enclosures
1752666 Members
5477 Online
108788 Solutions
New Discussion юеВ

Performance in XP12000

 
hboo
Frequent Advisor

Performance in XP12000

Hi
Somebody knows how many IO/s and MB/s is still inside a good performance parameters in an XP12000 array disk.
The array has 16 enable ports.
I have registered points of 1200 IO/s and 50 Mb/s in an single port and I wanted to know if this even is not a point of attention
Thanks
5 REPLIES 5
Sameer_Nirmal
Honored Contributor

Re: Performance in XP12000

The overall performance parameters for XP12000 could be found at

http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?objectID=c00241405тМй=en&cc=us&taskId=101&prodSeriesId=436460&prodTypeId=12169


Single port, performance details are at

http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?objectID=c00605822тМй=en&cc=us&taskId=101&prodSeriesId=436460&prodTypeId=12169

The above doc shows that maximum single port throughput is 255 Mbytes/sec. A Fibre Channel CHIP pair with a maximum single port throughput of 350 Mbytes/sec is planned to be released in the future.
I have read somewhere that each CHIP microprocessor is capable of approximately 3K IOPS.

So looking at your stats, you are well below the limits as of now.
Peter Mattei
Honored Contributor

Re: Performance in XP12000

First of all you need to be aware that the 16 port CHIP has a total of 8 microprocessors each serving 2 port.
The performance relevant part is the microprocessor (mp)!
One mp can do up to 26k read 8kB IOPs from cache and more than 3000 8kB IOPs read or write IOPs from/to the backend (cache avoidance)
If you are having 1200 8kB IOPs per mp your fine.
The MB/sec depend on blocksize as well.
One 2Gbps port can do up to 190MB/s with 64kB blocks. Two ports on the same mp can do 260MB/s.
So you need to know what you are looking at!

Best is if you have performance advisor installed where you can also look at the mp utilization which is very important.

Note that the values above are to be expected in simplex mode that means no Continuous Access involved!

Cheers
Peter
I love storage
Alzhy
Honored Contributor

Re: Performance in XP12000

Hayse,

It depends how many LUNs you have hanging off that single port (HBA) and how you built your storage on your OS and volume manager.

In my experience, I've realized much much better IOPS and MBps than what you have with XP Storage hanging off one Fibre-Channel/HBA. If your test dealt with testing on a single LUN, then that will be within the realm of my expectations as well.. but try striping say a set of XP LUNs accross the ACPs. Even if all are presented on a single FC port, your performance stats will be up there...

Hakuna Matata.
Peter Mattei
Honored Contributor

Re: Performance in XP12000

Nelson is right

The XP12000 has a huge performance potential but in order to take advantage of it you need to spread the load across all available resources.
This starts with the LDEVs and Array Groups, goes through the ACPs and then the CHIPs.

Depending on your OSes there are different best practices achieving that.

With HP-UX you want to have quite a number of LUNs/LDEVs spread across all available Array Groups and ACPs and put it in a volume group.

Using windows you can take advantage of building 2 or 4 striped groups of RAID5 (7&1) Array groups. Here you get LDEVs/LUNs that are striped across 14 or even 28 physical disks witch give you great single LUN performance.

But you only asked about CHIP performance ....

One important thing with CHIP performance: The XP CHIP microprocessor tries to optimize performance by reading ahead if necessary. This only works if the same stream is going through the same CHIP port! So best is setting multipathing behavior to preferred path and not round robin or so!

Cheers
Peter
I love storage
sivam
Advisor

Re: Performance in XP12000

AS per the XP Performance is concerned (16 ports) the load has to be distributed to all the active ports.

In a 2 gig HBA maximum of 2.5 TB per chip port (per host hba) can be allocated. This is maximum.

When you consider the thumb rule you need to distibute the load across all the chip ports.

Sivam /CVA