San Speed question

[rp4643]

I have hosts with 1Gb Hba,s, That connect to a Mcdata switch with 2Gb HBA's installed. I use a SVC with 2Gb Hba's and the backend is a ds8300 with 4Gb Hba's. I have TPC so I can monitor I/O performance. What should I being seeing for I/O R/W speed at the Mcdata? Is the 1Gb Hba bottlenecking me?

BTW I do run mutlipathing for redundancy using 2 Fc instead of 1. Would 4 provide a gain?

 

[KB]

It depends. The IO load is determined by the host not the backend disk or SVC. Unless your host machines are multi-cpu monsters it is not likely they are going to be able to generate a load that is capable of saturating a 1 GB HBA. In general it is the IOs per second that will affect the backend disk and that effect will show up as increasing IO response time. What is TPC telling you? Are you generating enough IO load to take your HBAs to full capacity?

 

[rp4643]

Reply to KB

The hosts do have several CPU's assigned on the Regatta's. I have no way of seeing how much load the generate. well not true, they are running on AIX 5.3 so I could run topaz or nmon and see this figure. We have a redundant SAN in that we are running 2 HBA ports per host. I do not have TPC data running only fabric and disk.

Im not sure what TPC is telling me at the Mcdata?

 

[KB]

OK. Regatta's were pretty healthy - they only have 1 GB HBAs? Which version of TPC are you using? If you have TPC Standard edition or disk and fabric, you should be able to get performance stats from the ports on the McData that are connected to the hosts. We should also be able to look at the ports related to the SVC and the disk from the TPC disk performance section. Are the hosts connected to the SVC or are they direct to the DS8300?

 

[rp4643]

KB-Answers

Yes the Reaggata have only 1GB Hba's, 2 assigned per Lpar(Fcs0 and FCS1), I think Our Squadrons are running 2 GB Hba's. Im using TPC 3.1.3 . Full Edition but not running data on Lpars, Havent got there yet. So Im seeing performance for the Mcdata, SVC(4 nodes) and DS8300 right now. Im running a week baseline right now. We do tons of Oracle transaction samoung other Applications on the Lpars, 4 Lpars are used for ERP and run RAC.

All hosts and the SVC's hook to the Mcdata's 4 FC's per SVC 2 going to one mcdata and 2 to the other. We Have 2 Mcdata (6140's) for each FC, So FCS0 uses one mcdata and FCS1 uses another. This is for redundancy. I have zoned the Ds8300 HBA,s with the SVC hba's with the LPAR Hba's for each Mcdata. We have a 4 node SVC environment with 2 I/o groups.

The DS8300 has 4GB HBA's.

 

[KB]

Collect your baseline data then let us know what you see.