Originally posted by: MYAU_Libor_Zurek

Hello.

aix 7.1 on P7, we observe strange behavior of tape hba (in this case meaning dual-port HBA which has it's ports zoned with tape devices only).

The strange thing is, it seems to have transmit throughput limited to about 220MB/s, but at adapter level, not port level (i.e., one port or the other can be using all of this throughput, although most of time they are running at 105MB/s on each port. Considering that SAN behind each port is different fabric, it's hard to imagine this limitation being in SAN.

Ports are 8Gbit, so should be capable of ~800MB/s each. On SAN, they appear correctly as 8Gbit, and there are no port errors, SAN is monitoried for various thresholds and there doesn't seem anything wrong (and QOS is not enabled, although that wouldn't match the behaviour here anyway).

Zoned with the hba ports are bunch of tape drives. Although drives are dual-port, they are only connected through one port (so, each AIX port can only see subset of drives, connected to particular fabric).

I'm not the AIX guy, but admin of TSM Server running on this system.

We have asked UNIX guys to open HW call for AIX/Atape, and we have also opened TSM Software call.

However if someone would have some immediate idea where to point/what to check, it would be most welcome.

For completion purposes, this system is in cluster with another system in other location. The other system is performing better (though perhaps also not as expected), but there are some differences (other system has adapters restricted to 4Gbit, due to physical link limitations). Otherwise systems should be same as far as I know.

EDIT:

I suppose it's also goot to mention that this behavior seems consistent almost all of the time, regardless of nature of the operations (i.e., backup data flowing through LAN, TSM Server database backup (Disk SAN), moves of data between tapes(backup copy from one tape to another, move from one tape to another). Among observed results are differences in TSM Server database backup ranging between 5 min(corresponding to max drive capability) to 50 min. So the only common factor here is the HBA. Also, there are no errors anywhere (we looked) - not on AIX, not in TSM, not on SAN.

Originally posted by: jklotz

What do you have in the output of the command : fcstat fcsX (fcsX being your FC adapter) ?

You should see something like this about port speed  :

# fcstat fcs0

FIBRE CHANNEL STATISTICS REPORT: fcs0
[...]
Class of Service: 3
Port Speed (supported): 4 GBIT
Port Speed (running):   4 GBIT

[...]

I'm not aware of any limitation of this kind at the HBA level, main factors would usually imply a physical problem in the link or in the SFP connected to the switch port (I observed some improved behaviours by just unplugging and plugging back a link in a switch, or several systems trying to access the target at the same time.

If possible, you might zone the ports to a SAN volume, and use the ndisk64 tool to make disk I/O test performance and see if you have differences when connected to tape devices.

regards