|
Otros manuales para este modelo:
Resumen del manual
In any system with IOAMEs or CLIMs, I/O traffic to any FCSA or CLIM is automatically distributed
across both ServerNet fabrics by SCS (T8456). This provides both fault tolerance and load balancing.
SCS automatically rebalances the load after failure and restoration of a ServerNet fabric.
However, the module number of an FCSA adapter does imply an affinity for a specific ServerNet
switch (module 2 = X fabric, module 3 = Y fabric) for FCSA reset processing, so the four paths to a
disk volume should be configured to use both fabrics as described above for S-series I/O. The
convention for identifying the location of a CLIM always uses a ServerNet connection point with
module=2. This does not indicate an affinity between CLIMs and the X fabric.
The connection points of a CLIM to an NS-series P-switch are identified by group, module, slot, and
port. The slot number identifies a FRU which holds four port connectors. For best fault tolerance, the -P
and -B paths to a disk volume should not use CLIMs connected to the same slot number. The same is
true for the -M and -MB paths and the -P and -M paths.
The connection points of a CLIM to an NB-series ServerNet switch are identified by group, module, slot,
port, and fiber. The port number identifies a connector which has a 1-to-4 splitter cable connected to it.
For best fault tolerance, the -P and -B paths to a disk volume should not use CLIMs connected to the same
port number. The same is true for the -M and -MB paths and the -P and -M paths.
In an S-series system, the four low-numbered processor enclosures (groups 1 through 4) are directly
connected to each other and each I/O enclosure is directly connected to only one processor
enclosure. The four high-numbered processor enclosures are only connected to one other processor
enclosure each (1-5, 2-6, 3-7, and 4-8). The loss of a low-numbered processor enclosure causes the
isolation of its directly connected I/O enclosures and the directly connected high-numbered processor
enclosure and its directly connected I/O enclosures.
Adapters, SACs, CLIMs, and configuration limits
In S-series systems running G06.04 or older RVUs, all the adapters used by an IOP had to be in the
same S-series enclosure and that enclosure had to be in the same topology branch as the IOP. A
"topology branch" is an S-series processor enclosure and all of its directly connected to I/O
enclosures. Starting with the G06.05 RVU, this constraint was relaxed so that the adapters for an IOP
could be spread across multiple enclosures within one topology branch. Starting with the G06.23
RVU, and in all H-series and J-series RVUs, the adapters and IOP can be spread across the entire
system. But spreading the adapters across multiple topology branches in an S-series system increases
the ServerNet path length, so a reasonable compromise is to put the adapters in the same topology
branch(es) as the IOP.
In systems running any G-series RVU and in systems running H06.07 or older RVUs, only two
processors can share access to a SCSI SAC, so all of the devices connected to a SCSI SAC must have
their IOPs running in the same two processors. Starting with the H06.08 RVU, and in all J-series RVUs,
this constraint was relaxed so that eight processors can share access to a SCSI SAC at the same time.
With NS-series I/O connected to any system type, an IOP in any processor can use any FCSA
adapter. But there are some maximum configuration limits. In S-series systems running the G06.23 or
older RVUs, a total of 256 device paths can be configured to use an FCSA. Starting with the G06.24
RVU, and in all H-series and J-series RVUs, this constraint was relaxed so that each processor can
have up to 125 paths configured through each FCSA. Paths to all device types (disk, tape, and Open-
SCSI) are included in this limit.
To maximize the number of IOPs that can be configured in a processor, each IOP should use 4
different FCSAs for its 4 paths (-P, -B, -M, and -MB). Then each IOP only has one path through each
FCSA and 125 IOPs can be configured in the same processor. If more than 4 FCSAs are used, then
even more IOPs can be configured in the same processor, but available memory also limits the
5
...