Subscribe

read performance much worse than write?

Hi,

 

2-node FAS8040 with flash cache running in clustered ontap 8.3GA with 10GbE ports.

 

From a Linux client with single 10GbE connection, I ran several tests using iozone to obtain write/rewrite and read/reread performance numbers for an NFS mounted volume on NetApp. I notice all runs produce reads much less than writes. As an example,

 

kB            reclen    write       rewrite        read       reread      random read     random write   
8388608       4    2774799  2079447   111926  4616036  3701496           2193816
8388608       8    2963944  3444805   111376  5532988  5025833          2742466
8388608      16  3227913  3766890    110325  5083604  5953205        3247727
...

 

Notice for all record lengths, reads are in ~100MB/sec while writes are much higher, 2.7 - 3GB/sec.

 

1. I question the numbers for writes in the range of 3GB/sec which is excessively high even over 10Gbps link.

2. the 'read' numbers are way smaller than writes which does not make sense to me.

 

Could someone shed some lights on what I've missed?

 

Thanks,

 

Re: read performance much worse than write?

Hi,

 

the file you're testing with, should be much bigger than the RAM and cache in the box.

Otherwise you measure the cache throughput.

 

Marcus