ITEM: I5000L
hacmp questions
Question:
I have a model 520 with 3-670mb drives, 1 for rootvg, 2 for data. I want to
setup a HACMP environment.
(1) can I use a 3xx series as the second machine? what are the system requirements?
(2) what type of external disk do I need? 9334?
(3) do the terminals have to be ethernet or token-ring attached? will async work?
I want to setup a minimum system.
Response:
(1) According to salesman there is no restriction in using a 5xx series
with a 3xx series. This is the only restriction on choosing:
o The POWERserver 7013 (500 series models) should not be coupled with
POWERserver 7011 (200 series models).
(2) According to salesman here are the disk subsystems available:
The following table shows each type of media, which adapter is available for
that media, which disk enclosures are available for that adapter and which
disk drives and sizes are available for that enclosure. ONLY these
combinations are valid for HACMP/6000 Version 2.1. SCSI-2 Differential or
serial media is preferred wherever possible in place of SCSI SE media.
Max. \# of
Enclosures
Type Adapter Enclosure Disk Size(GB) per SCSI bus
----------- ------- --------- --------- -------- -------------
SCSI-1 SE 2835 7203 2300/7203 .355 5
2310/7203 .670 5
7204-320 2320/7204 .320 5
9334-500 2510 .670 1
2530 .857 1
2570 1.37 1
SCSI-2 SE 2410 7203 2320/7203 1.0 2
7204-001 001/7204 1.0 2
SCSI-2 Diff 2420 9334-011 2565 1.0 1
2585 2.0 1
9334-501 2565 1.0 1
2585 2.0 1
7135-110 2716 1.3 1
2720 2.0 1
7204-215 2585 2.0 3
Serial 6210 9333-010 3100 .857
3110 1.07
9333-500 3100 .857
3110 1.07
Serial 6211 9333-010 3100 .857
3110 1.07
9333-500 3100 .857
3110 1.07
Serial 6212 9333-010 3100 .857
3110 1.07
9333-500 3100 .857
3110 1.07
9333-011 3100 .857
3110 1.07
3120 2.0
9333-501 3100 .857
3110 1.07
3120 2.0
And here is the cabling required:
Disk Cabling for HACMP/6000:
For a given cable, any item listed in the "from" column can be connected to
any item in the "to" column. A serial bus can use any of the serial cables.
Serial bus cabling is not covered in the following tables.
SCSI-1/2 SINGLE ENDED CABLING
Len
Feature Type From To (m) Notes
--------- ----------- ------- -------- ---- ------------------------
3130 SCSI-1 SE 7203, 7203, 0.66
7204 7204
3130 SCSI-2 SE 7203, 7203, 0.66
7204 7204
2915 SCSI-1 SE Adapter 7203, 1.57 Passthru terminator
(2835) 7204 cable. Withdrawn from
market. Available as
RPQ \#8A0759.
2915 SCSI-1 SE Adapter 9334-500 1.48 Passthru terminator
(2835) cable. Withdrawn from
market. Available as
RPQ \#8A0759.
2914 SCSI-2 SE Adapter 7203, 1.57 Passthru terminator
(2410) 7204 cable. Withdrawn from
market. Available as
RPQ \#8A0758.
The configurations in the following table assume that the two processors are
at the end of the bus (just prior to each terminator) and all the storage
devices are connected to the bus between the processors.
SCSI-2 DIFFERENTIAL CABLING
Len
Feature From To (m) Notes
--------- ------- -------- ---- --------------------------------
2422 Adapter 2919, .765 Y-cable; base to adapter, long
(2420) 9334 cable, leg to 9334 cable or 7135-110
terminator interposer (2919), short leg is
terminated
none Y-cable self 0 Terminator, 50 pin, hi-density,
included when the Y-cable is
ordered.
2921/2923 Y-cable 9334-011 2921 - 4.75 m
2923 - 8.0 m
8.0 m cable must be paired with
the 4.75 m cable
2921 Y-cable 7504-215 4.75
2848 7204 7204 2
2931-2937 Y-cable 9334-501 2931 - 1.48 m
2933 - 2.38 m
2935 - 4.75 m
2937 - 8.0 m
8.0 m cable must be paired with
a 4.75 m or shorter cable
2919 Y-cable 7135-110 0 Cable interposer; connects 50
cable pin Y-cable to 68 pin 290X cable
for 7135-110
2901-2914 7135-110 2919 Connects 7135-110 array
controller to an interposer
(2919)
2901 - 0.6 m
2902 - 2.4 m
2905 - 4.5 m
2912 - 12 m
2914 - 14 m
12 m and 14 m cables must be
paired with a 0.6 m or 2.4 m
cable
The recommended 7135-110 bus configuration with only one 7135-110 is:
o Two controllers on the 7135-110; one active, one passive
o Connect the active controller to two processors
o Connect the passive controller to the same two processors
This yields two different SCSI buses, one with both processors connected to
the active controller and the second SCSI bus with both processors connected
to the passive controller. The RISC software manages access to the different
controllers and will switch controllers (from the active controller to the
passive controller) if the active controller fails; this occurs independently
of HACMP.
According to HACMP Planning and Installation Guide the only supported
media for concurrent access are the 7135 RAID unit, 9333 serial unit, and
9334 SCSI-2 DE unit.
The customer has several 670mb drives. He should be able to use the 9334-500
with the SCSI-1 SE drives even though SCSI-2 DE or serial is preferred.
(3) The terminals need to be able to switch between hosts for crash recovery.
There are serveral options (as listed by item Q606194 in ASKQ)
I can recommend three alternative means of getting asynchronous communication
services failed-over in an HACMP/6000 environment:
1. The first method uses IBM 3151s (or any other terminal that will
support the function) with the "Cartridge for Connectivity" option.
This option lets you connect the 3151 to two separate tty connec-
tions and "hot-key" between the sessions. If each of these sessions
is on a separate machine, you can hot-key in the event of a failure.
2. The second method is to switch at the adapter. If you are using a
64-port adapter, you run four RJ-45 triple-twisted-pair connections
to four concentrators that are in turn cabled to your devices,
you could switch that cable between two adapters on different
systems. In this way, either of the processors serve the term-
inals. The switch is not available from IBM, but any number of
catalogue-style vendors would be glad to provide one to you. An
improvement on this configuration is something we are working on
here at the Center with "Black Box". We are in the process of
having them build a box that we designed that would switch the
RJ-45 connection electronically. The command to switch would be
sent from the surviving machine in the cluster in a protocol defined
by the box, from a command executed as a part of the topchng.rc
script. When this configuration is completed, we will post some
pricing information to this data base -- probably in a "HOWTO AIX"
pricing information to this data base -- probably in a "HOWTO AIX"
item. In the mean time, if this method appeals to you, you might
ask Black Box or some other vendor to configure your connection
for you.
As for how the box would work, sometimes a picture is worth a thousand
words:
---------------------
| 16-Port |
| Concentrator |
---------------------
|
|
|--->|-----|
| A-B |
|-----|\<---
| | |RS-232
| | |
------------ | | | -------------
| Mach. A | | | |-->| Mach. B |
| | | | | |
| | | | | |
| 64-port | | | | 64-port |
| =====|\<-----------| |------->|===== |
| | | |
------------ -------------
3. The last method is probably the most trouble-free, and more directly
responds to your need for a Token Ring connection. A terminal server
essentially is a network client that is smart enough to manage tty
devices. In almost all cases, terminal servers attach to Ethernet
networks -- not Token Ring. I have only ever heard of one producer
of terminal servers that connects to Token Ring, but I'm afraid I
don't know what their name is. I believe that they are an IBM bus-
iness partner. Don't abandon hope, however. An alternative approach
is to use a smart client that can attach to Token Ring and also to
async terminals. An example of this would be our own 220. Although
it is priced comparably to a terminal server (a bit more when you
consider the 64-port card and Token Ring adapter), it is much more
powerful, working with a full-featured operating system. This is
particularly valuable with HACMP, because it will allow the Cluster
Information Services (CLINFO) daemon to run without porting. Once
you configure IP Address Takeover, clinfo will purge the client's
ARP cache so that the takeover machine resumes communication rel-
atively transparently.
IBM does not currently (4/19/94) make a terminal server. In choosing a
terminal server (according to HACMP Planning and Installation Guide) you
need to consider:
1. Can you update the terminal server's ARP cache? The terminal server must
comply with the TCP/IP protocol, including telnet. If you will have
modems for dial-in users, it must also support SLIP.
2. Is the terminal server programmable, or does it need manual intervention
when a cluster event happens?
3. Can you have a control file from the cluster node to the terminal server
that updates/handles cluster events' effects on clients?
If your terminal server does not meet these guidelines, you should choose the
Hardware Address swapping option when configuring the cluster environment.
Other considerations are summarized in question Q605036 on ASKQ:
QUESTION:
If I plan to install HA/6000 and run in mode 1 (standby) must
I license two copies of HA/6000 (one on the primary Risc machine
and one on the standby machine)?
2. Must I use two Ethernet adapters or does two adapters
allow me backup for my network should it go down?
3. Since the application is critical I plan to suggest mirroring.
Can I mirror the 9333 disk drive with another?
4. I would like to recover from a Risc System/6000 failure as quickly
4. I would like to recover from a Risc System/6000 failure as quickly
as possible. What hardware should I recommend? I am considering 9333
versus 9334 (SCSI). I am also looking into high speed fiber between
the two Risc Systems. Would the fiber improve takeover time or is the
takeover time just dependent on the scripts and the takeover tasks?
Regarding \#2 above do I need to wire two distinct lans for network
redundancy? What terminal servers are recommended to facilitate
moving to the second Risc System in case of failure of the primary?
---------- ---------- ---------- --------- ---------- ----------
A: 1) You will need to buy one HACMP/6000 license for each machine
that will be run the HA server code.
2) In order to have network redundancy, you will need to wire two
separate LANs and have a minimum of two network adapters per
separate LANs and have a minimum of two network adapters per
machine. This will insure single point of failure protection.
3) Yes, you can mirror a 9333 with a second 9333. You can also
mirror the disks internally within a 9333.
4) The 9333 is a better choice than the 9334 as far as recovery
time. Purchasing a fiber link is probably a waste. Most of
the time during recovery is during the configuration on the
backup server. Fiber might be a couple of milliseconds faster
in determining a fault with the primary server.
5) You could use a 220 as a terminal server. This would allow you
to run clinfo so that the 220 would automatically switch over
when the primary server failed.
Customer had more questions on HACMP:
(1) Does HACMP takeover immediately after failure? How does it do it?
(2) Can I have something else on the SCSI bus besides the 9334-500?
(3) What type of ethernet requirements do I need?
Response:
Called Jeff Grubb to get more information. The 9334-500 requires a \#2835
scsi adapter for each machine. It also requires a scsi PTT cable which
is available as an RPQ \#8A0759.
(1) When the primary server goes down HACMP allows the 2nd backup to switch
ip addresses with the primary, remount all the filesystems, mount any
nfs filesystems, and reinitiate a getty to all terminals within
30 - 300 seconds.
In addition, HACMP can be setup in a type of mutual takeover that
will allow you to run secondary applications on the backup machine.
In the event of failure all applications would be run on the backup.
(2) The 9334 can be the only device on the bus.
(3) You will need two ethernet controllers for each system. One to connect
to the network and one to connect between the two machines.
There is a publication that can tell you more about HACMP capabilities.
It is publication SC23-2595 - HACMP System Overview.
Support Line: hacmp questions ITEM: I5000L
Dated: April 1994 Category: N/A
This HTML file was generated 99/06/24~13:30:45
Comments or suggestions?
Contact us