Showing results for 
Search instead for 
Do you mean 

EVA performance

SOLVED
Go to Solution
Occasional Advisor

EVA performance

Hi,

were trying to plan how our new EVA disk stack is going to perform on our HP-UX servers. Can anyone provide a way to test it ?
12 REPLIES
Honored Contributor

Re: EVA performance

Just use time raw dd. Create an lvol, make it at least 1000MB (1GB) then run some dd's on it to see how fast it is;

time dd if=/dev/vgXX/rlvolYY of=/dev/null bs=1024k count=1000

The resulting time, take 1000 (MB) / time in seconds to get the transfer rate.
Im from Palmerston North, New Zealand, but somehow ended up in London...
Honored Contributor

Re: EVA performance

Hi,
Performance tests on a large disk system is a rather complex process as it is several factors that may affect your application performance (e.g. throughput, random reads/writes sequential read/writes, cache memory etc.). In most cases, IO performance (IOPS) is more interesting then throughput.

Maybe you can have some use of the Postmark benchmark:

http://www.netapp.com/tech_library/3022.html
Occasional Advisor

Re: EVA performance

Stephen,

why do you use rlvol and not lvol for dd ?
Honored Contributor

Re: EVA performance

rlvol bypasses the unix buffer cache so not only is it faster but you get a more accurate result (if you use lvol for dd the data may be in cache which will affect the response time giving you a faster result than you should get).
Im from Palmerston North, New Zealand, but somehow ended up in London...
Occasional Advisor

Re: EVA performance

my time dd on my EVA lvol takes exactly 8 seconds all the time, which is only 125MB/s. I thought it would be much faster than this. What am I doing wrong ?
Honored Contributor

Re: EVA performance

125MB/s exactly is the speed of the individual fibre connected drives on your EVA. Sounds like you are only using 1 disk on the EVA in your disk group. Create a bigger disk group, use something like a minimum of 12 disks (max 32). Then recreate your vdisk and then vg and lvol and try again. You should get around 160+ MB/s.
Im from Palmerston North, New Zealand, but somehow ended up in London...
Occasional Advisor

Re: EVA performance

Thanks Stephen,

I can now get over 140 MB/s which is an improvement, but not 160+MB/s. What else must do ?
Honored Contributor

Re: EVA performance

Stripe your lvol when you create it also. Add in > 1 lun from the EVA to your HP server, then add these luns into 1 VG, then lvcreate -i 2 -I 64 and rerun your dd, this should boost to 160MB/s +
Im from Palmerston North, New Zealand, but somehow ended up in London...
Occasional Advisor

Re: EVA performance

Is there anything else I can do to boost performance ?
Honored Contributor

Re: EVA performance

You should use >1 fibre card to connect to your EVA. Then you can present LUNS down both channels to your HP server and then create your lvol down both channels and this will boost performance to around 230+MB/s on a striped lvol (for a single dd test).

Im from Palmerston North, New Zealand, but somehow ended up in London...
Honored Contributor

Re: EVA performance

A single dd test is not the best way to effectively test performance either. There is a tool from HP called stkio.hp which is designed to test EVA performance by running multiple transfers simeltaneously (eg. running multiple dd's at the same time). With this tool you will see performance with say 2x 2GB fibres and 5 threads jump from 230+MB/s to around 330MB/s. Thats the best we can get. I think you have to ask HP for this tool though.
Im from Palmerston North, New Zealand, but somehow ended up in London...
Honored Contributor

Re: EVA performance

oops, forgot to mention, my last reply was with 50GB luns from the EVA, with 100GB luns the performance jumped from 330+MB/s to 378MB/s - thats the best we got after experimenting with different lun sizes.
Im from Palmerston North, New Zealand, but somehow ended up in London...
//Add this to "OnDomLoad" event