Original text:
http://vmfocus.com/2013/08/22/3par-storeserv-zoning-best-practice-guide/
This is an excellent guide which has been written by
Gareth Hogarth
who has recently implemented a 3PAR StoreServ and was concerned about
the lack of information from HP in relation to zoning. Being a ‘stand
up guy’ Gareth decided to perform a lot of research and has put together
the ’3PAR StoreServ Zoning Best Practice Guide’ below.
This article focuses on zoning best practices for the StoreServ 7400
(4 node array), but can also be applied to all StoreServ models
including the StoreServ 10800 8-node monster.
3PAR StoreServ Zoning Best Practice Guide
Having worked on a few of these, I found that a single document on
StoreServ zoning BP doesn’t really exist. There also appears to be
conflicting arguments on whether to use Single Initiator – Multiple
Target zoning or Single Initiator – Single Target zoning.
The information herein can be used as a guideline for all 3PAR supported
host presentation types (VMware, Windows, HPUX, Oracle Linux, Solaris
etc…).
Disclaimer: Please note that this is based
on my investigation, engaging with HP Storage Architects and
Implementation Engineers. Several support cases were opened in order to
gain a better understanding of what is & isn’t supported. HP
recommendations change all the time, therefore it’s always best to speak
with HP or your fabric vendor to ensure you are following latest
guidelines or if you need further clarification.
Right, let’s start off with
Fabric Connectivity
In terms of host connectivity options the StoreServ 7000 (specifically the 7400) provides us with the following:
- 4x built-in 8 Gb/s Fibre Channel ports per node pair.
- Optional 8 Gb/s Quad Port Fibre Channel HBA (Host Bus Adapter) per node (we will be focusing on this configuration option).
- Optional 10 Gb/s Dual Port FCOE (Fibre Channel over Ethernet) converged network adapter per node.
StoreServ target ports are identified in the following manner
Node:Slot:Port.
StoreServ target ports located on the on-board HBA’s will always
assume the slot identity of 1, respectively StoreServ targets ports
located on the optional expansion slot will always assume the identity
of slot 2.
StoreServ nodes are grouped in pairs, it’s important to pay
particular attention to this when zoning host initiators (server HBA
ports) to the StoreServ Target ports.
Recommendations
- Each HP 3PAR StoreServ node should be connected to two fabric switches.
- Ports of the same pair of nodes with the same ID (value) should be connected to the same fabric.
- General rule – odd ports should be connect to fabric 1 and even ports should be connected to fabric 2.
Figure 1a below identifies physical cabling techniques, mitigating
against single points of failure using a minimum of two fabric switches,
which are separated from each other.
The example below illustrates StoreServ nodes with supplementary quad port HBA’s:
Moving on to
Port Persistence
As already covered by Craig in this
blog post,
a host port would be connected and zoned on the fabric switch via one
initiator (host HBA port) to one HP 3PAR StoreServ target port
(one-to-one zoning). The pre-designated HP 3PAR StoreServ backup port
must be connected to the same fabric as its partner node port.
It is best practice that a given host port sees a single I/O path to
HP 3PAR StoreServ. As an option, a backup port can be zoned to the same
host port as the primary port, which would result in the host port
seeing two I/O paths to the HP 3PAR StoreServ system. This would also
result in the configuration where a HP 3PAR StoreServ port can serve as
the primary port for a given host port(s) and backup port for host
port(s) connected to its partner node port.
Persistent ports leverage SAN fabric NPIV functionality (N_Port ID
Virtualization) for transparent migration of a host’s connection, to a
predefined partner port on the HP 3PAR StoreServ array during software
upgrades or node failure.
One of the ways this is accomplished is by having a predefined host
facing port on the 3PAR StoreServ array, so that in the event of upgrade
(node shutdown) or node down status the partner port will assume the
identity of its partner port. The whole process is transparent to the
host. When the node returns to normal I/O is failed back to the original
target port.
Although unconfirmed I have heard that in in future releases of
Inform OS we will get this level of protection at the port level.
Essentially for this to work Port Persistence requires that
corresponding ‘native’ and ‘guest’ StoreServ ports on a node pair, be
connected to the same fibre channel fabric.
Requirements for
3PAR Port Persistence:
- The same host ports on the host facing HBA’s in the nodes in a node pair must be connected to the same fabric switch.
- The host facing ports must be set to target mode.
- The host facing ports must be configured for point-to-point connections.
- The Fibre Channel fabric must support NPIV and have NPIV enabled on the switch ports.
Checking and enabling NPIV
Brocade Fabric OS (ensure you have the appropriate license which enables NPIV)
admin> portcfgshow ‘port#’
If the NPIV capability is enabled, the results of the
portcfgshow command will identify this, i.e
NPIV capability ON.
If the NPIV capability is not enabled, you can turn it on with the following command:
admin> portCfgNPIVPort ‘port#’ 1 (1 = on, 0 = off)
Cisco MDS Series Switches
fabSwitch # conf t
fabSwitch(config) # feature npiv (Enables NPIV for all VSANs on the switch)
QLogic SANbox 3800, 5000 and 9000 Switches
Don’t require a license, it’s enabled by default (just ensure you are using firmware version 6.8.0.0.3 or above).
Now let’s cover
Switch Zoning (Fibre Channel)
SAN zoning is used to logically group hosts and storage devices
together in a physical SAN, so that authorised devices can only
communicate with each other if they are in the same SAN zone.
The function of zoning is to:
- Restrict access so that hosts can only see the data they are authorised to see.
- Prevent RSCN (Registered State Change Notification) broadcasts.
What are ‘RSCNs’ ? RSCNs are a feature of fabric switches. It’s a
service of the fabric that notifies devices of changes on the state of
other attached devices. For example if a device is reset, removed or
otherwise undergoes a significant change in status.
These broadcasts are made to all members in the configured SAN zone.
As hosts and storage targets can be grouped in a zone its best practice
to reduce the impact of these types of broadcasts (
Note: an
argument against RSCN’s causing issues in zoning tables is that newer
HBA’s do a good job limiting the impact of these types of broadcasts).
Nevertheless, I prefer limiting the number of initiators and targets in
a fabric zone to a minimum.
Zoning Types
- Domain, Port zoning uses switch domain id’s and port numbers to define zones.
- Port World Wide Name or pWWN zoning uses port World Wide Names to
define zones. Every port on a HBA has a unique pWWN. (A host HBA
comprises of a – nWWN & a pWWN, the nWWN refers to the whole device
whereas the pWWN refers to the individual port.
The preferred zoning unit for the 3PAR StoreServ is pWWN. If you are
currently using Domain, Port migrating to pWWN is very easy. Simply
create new zones based on the pWWN of the host and the pWWN of the
storage target, add these new zones to your fabric switches, zoning-out
the references to Domain, Port for that respective HBA port. Some fabric
vendor’s support mixing both Domain, Port and pWWN in the same zone. I
prefer using one or the other explicitly.
The following command outputs the StoreServ ports and partner ports, which can be used to identify the node pWWN’s for zoning.
3PAR01 cli%
showport
HP 3PAR StoreServ supports the following zoning configurations:
- Single initiator – Single Target per zone (recommended)
- Single initiator – Multiple Targets per zone
Use Single Initiator – Single Target per zone over Single Initiator
Multiple Targets per zone to reduce RSCN’s as previously discussed.
At the time of writing, HP 3PAR OS implementation documentation
references Single Initiator Multiple Targets as the recommended zoning
type. However, when I queried this I was directed to use Single
Initiator – Single Target Zoning. HP support pointed me in the
direction of this document which identifies Single Initiator – Single
Target zoning as best practice:
http://www8.hp.com/h20195/v2/GetDocument.aspx?docname=4AA4-4545ENW
HP will support Single Initiator – Multiple Target, but you should
not have a single host initiator attached to more than two StoreServ
target ports!
Host port WWN’s should be zoned in partner pairs. For example if a
host is zoned to node port 0:2:1, then it should be zoned to node port
1:2:1 (I’m speculating here, but I guess this is because controller
nodes mirror cache I/O, so that in the event of node failure write
operations in cache are not lost – hence we zone in node pairs and not
across nodes from different pairs).
After you have zoned the host pWWN to the StoreServ node pWWN, you can use the 3PAR CLI
showhost command
to ensure that each host initiator is zoned to the correct StoreServ
target ports (ensuring initiators go to different targets over different
fabrics).
Figure 1b represents a staggered approach where you would have odd
numbered VMware hosts connecting to nodes 0 & 1, and even numbered
hosts connecting to nodes 2 & 3 (
Note: currently the StoreServ is designed to tolerate a single node failure only, this includes the 8-node StoreServ 10800 array).
The example depicts Single Initiator – Single Target zoning, so a
host with two HBA ports connecting over two fabrics will have a total of
four zones (two per fabric). In case you were wondering the maximum
allowed is eight (also known as fan-in limitation which is four per
fabric).
Here are some
additional points to be aware of
Fan-in/Fan-out ratios:
- Fan-in refers to a host server port connected to several HP 3PAR storage ports via Fibre Channel switch.
- Fan-out refers to the HP 3PAR StoreServ storage port that is connected to more than one host HBA port via Fibre Channel switch.
Note: Fan-in over subscription represents the flow of data in
terms of client initiator to StoreServ target ports. HP/3PAR
documentation states that a maximum of four HP 3PAR storage system ports
can fan-in to a single host server port (if you are thinking
great, I’ll connect my VMware host to 8 ports [four per fabric] think
again. Using this approach when you have hundreds of hosts can quickly
reach the maximum StoreServ port connection limitation which is 64!)
it’s just not necessary.
StoreServ Target Port Maximums (As per 3PAR InForm OS 3.1.1 please observe the following):
- Maximum of 16 hosts initiators per 2Gb HP 3PAR StoreServ Storage Port
- Maximum of 32 hosts initiators per 4Gb HP 3PAR StoreServ Storage Port
- Maximum of 32 hosts initiators per 8Gb HP 3PAR StoreServ Storage Port
- Maximum total of 1,024 host initiators per HP 3PAR StoreServ Storage System
HP documentation states that these recommendations are guidelines,
adding more than the recommended hosts should only be attempted, when
the total expected workload is calculated and shown not to overrun
either the queue depth or throughput of the StoreServ node port.
Note: StoreServ storage ports irrespective of speed, will
negotiate at the lowest speed of the supporting fabric switch (keep this
in mind when calculating the number of host connections).
The following focuses on changing the target port queue depth on a VMware ESX environment.
The default setting for target port queue depth on the ESX host can
be modified to ensure that the total workload of all servers will
not overrun the total queue depth of the target HP StoreServ system
port. The method endorsed by HP is to limit the queue depth on a
per-target basis. This recommendation comes from limiting the number of
outstanding commands on a target (HP 3PAR StoreServ system port), per
ESX host.
The following values can be set on the HBA running VMware vSphere.
These values limit the total number of outstanding commands the
operating system routes to one target port:
- For Emulex HBA target throttle = tgt_queue_depth
- For Qlogic HBA target throttle = ql2xmaxqdepth
- For Brocade HBA target throttle =
bfa_lun_queue_depth
(
Note: for instructions on how to change these values follow
VMware KB1267, these values are also adjustable on Linux Redhat & Solaris).
The Formula used to calculate these values is as follows:
(3PAR port queue depth [see below]) / (total number of ESX severs attached) = recommended value
The I/O queue depth for each HP 3PAR StoreServ storage system HBA mode is shown below:
Note: The I/O queues are shared among the connected host server HBA ports on a first come first serve basis.
| HP 3PAR StoreServ Storage HBA |
I/O queue depth values |
| Qlogic 2Gb |
497 |
| LSI 2Gb |
510 |
| Emulex 4Gb |
959 |
| HP 3PAR HBA 4Gb |
1638 |
| HP 3PAR HBA 8Gb |
3276 |
Well, hopefully you found the above information useful. Here is a
high level summary of what we have discussed:
- Identify and enable NPIV on your fabric switches (Fibre Channel only
feature – NPIV-Port Persistence is not present in iSCSI environments)
- Use Single Initiator -> Single Target zoning (HP will support
Single Initiator – Multiple Target, but you should not have a single
host initiator attached to more than two StoreServ target ports).
- A maximum of four HP 3PAR Storage System ports can fan-in to a single host server port.
- Zoning should be done using pWWN. You should not use switch port/Domain ID or nWWN.
- A host (non-hypervisors) should be zoned with a minimum of two ports
from the two nodes of the same pair. In addition, the ports from a host
zoning should be mirrored across nodes.
- Hosts need to be zoned to node pairs. For example, zoned to nodes 0
and 1 or to nodes 2 and 3. Hosts should NOT be zoned to non-mirrored
nodes such as 0 and 3.
- When using hypervisors, avoid connecting more than 16 initiators per 4 Gb/s port or more than 32 initiators per 8 Gb/s port.
- Each HP 3PAR StoreServ system has a maximum number of initiators supported, that depends on the model and configuration.
- A single HBA zoned with two FC ports will be counted as two
initiators. A host with two HBA’s, each zoned with two ports, will count
as four initiators.
- In order to keep the number of initiators below the maximum supported value, use the following recommendations:
- Hypervisors: four paths maximum.
- Other hosts (non-hypervisors): two paths to two different nodes of the same port pairs.
- Hypervisors can be zoned to four different nodes but the hypervisor
HBAs must be zoned to the same Host Port on HBAs in the nodes for each
Node Pair.
Reference Documents
HP SAN Design Reference Zoning Recommendations
HP 3PAR InForm® OS 3.1.1 Concepts Guide
The HP 3PAR Architecture
HP UX 3PAR Implementation Guide
HP 3PAR Red Hat Enterprise Linux and Oracle Linux Implementation Guide
HP 3PAR VMware ESX Implementation Guide
HP 3PAR StoreServ Storage and VMware vSphere 5 best practices
HP 3PAR Windows Server 2012, Server 2008 Implementation Guide
HP Brocade Secure Zoning Best Practises
HP 3PAR Peer Persistence Whitepaper
An introduction to HP 3PAR StoreServ for the EVA Administrator
Building SANs with Brocade Fabric Switches by Syngress
