Current config of G1000:
Pool GOLD - Disk PG RAID Pool SILVER - Disk PG RAID
FMD 1 R5(7+1) FMD 1 R5(7+1)
SAS 10K 34 R6(6+2) SAS 10K 14 R6(6+2)
SAS 7.2K 4 R6(6+2) SAS 7.2K 12 R6(6+2)
We grouped our AIX hosts to 5-7 hosts per 4 storage ports. (So, each host accesses 4 storage ports thru 2 fabrics).This design was done with Hitachi's help.
For this particular case, we have 5 AIX hosts zoned to the 4 storage ports (CL1-E, CL2-G, CL1-G, CL2-E).
Whenever host A transfers at high throughput (using large IO size and it's thruput capacity), the response time for all other hosts is affected (see attach doc). I have noticed that this issue occurs only when any AIX host transfers using large IO sizes >150KB on any storage port.
All hosts have 2x4Gb ports and the G1000 ports are 8Gb. The switch shows that the host is obtaining 4Gb/s and the storage port is at 50% (which matches that of the host).
Question: Why are all other hosts on the same storage ports affected? I was under the impression that the pool can not supply when IO size are large (> 150KB), but the other hosts in the same pool and on other storage ports do not seem to be affected. Can it be the ports on the G1000 that create the delay?