hi,

first , i migrated all data on the DMX3, so the DMX1000 is empty.

i deleted all devices , created 16 devices ( R-5 about 25Gb, hve about 8 Gb).
with these 16 devices i created a M16 concatnated , after a M16 striped.

i read a large file ( about 12 Mb ), with a troughput of 14-18 MB ...
if i read 2 large files in Parallel mode , i obtained a througput of 33-38 MB.
2 files in sequential mode , 14-18MB

i tested with IOmeter with this M16 ( striped ), i also allocated these 16 devices to form a volume under windows ( with dynamic disks ... sorry).
align 64 and 32k block size file system.

first test : one worker with 16k ios -> 1200-1300 IO/s (throughput20-22MB )
32K ios -> 770 IOs/s ( 24 MB )

second test : 4 workers on the same device
16K io -> 3500 IOs /sec ( 56MB
32L io -> 2300 IOs/sec ( about 70 MB)

a symstat -type request on the device shows about 90-100% read hit.

could you explain me why i can't obtain a throughput of 56MB with one worker like 4 workers ?
same with throughput of 2 files for one file ?

i read some doc and found the terms of Message mode Parallel mode .i'm on the right way?


test on dmx3 are a few better, but 4 workers on the same device have a throughput of 100 MB/sec, and one borker about 25-30 MB/sec.

enguinity are 5771 and 5772.
DMX1000 has 2 DA, DMX3 is a 3500 model
i did my best to create the meta on the maximum of DA:port

EMC support found nothing, the dmx1000 and dmx3 are fine

thanks

stephane