HACMP PROPOSED CONFIGURATION AND QUESTIONS!
ITEM: RTA000027792
QUESTION:
I am in the process of putting together a high availability
solution for a customer. Please read the following
proposed solution and address the questions I have after
it.
The configuration is to have two RISC/6000's configured in
a mode 2 HACMP setup. The first RISC will be a 980 with
14 GB of 1.07 GB serial drives. I will max out the driver
drawers in the base 980 rack, then add a 7202-900 and put
the additional disks in the drawers there. This 980 will
have 4 64 port adapters to start. It will also have the
optical serial link and dual ethernet cards. The
application running on this 980 will be the Advanced Pick
database (from Pick Systems). This application writes to a
raw logical volume and bypasses the standard JFS
filesystem.
The other RISC/6000 is a model 580. This will have 4 64
port adapters with no concentrators attached. In the case
of the 980 failing, the system operator will detach the
concentrators from the 980 64 port adapters and just plug
the concentrators into the model 580 64 port adapters.
This 580 will be running Netware for AIX. The Netware
clients will be attached across the dual ethernet cards
which will also be attached to the 980. The 580 will have
a serial optical connection to the 980. The 580 will also
be connected to the 1.07 GB serial drives which make up the
14GB I described in the 980 config above.
I would like the 580 to 'backup' the 980 and it's Pick
application with ascii terminals. I would like the 980 to
'backup' the 580 and it's Netware/6000 application with
ethernet attached clients. I would like to set this whole
thing up using the hacmp product. It looks pretty possible
so far, but I do have some concerns and questions.
My questions are as follows:
1) Please comment on the feasability of this solution.
Specifically address:
A) whether Netware/6000 will be able to work in/and
take advantage of an hacmp configuration.
B) whether the 580 can attach to the serial disk in
both the 7202-900 as well as the disk drawers in the
base 980 rack
C) whether just simply swapping the 64port
controller concentrators is the best solution, or
would terminal servers be better? Why?
D) any other issues you see here that I should be
aware of?
2) A) In an HACMP solution, when the 'failover' process
occurs what happens to the data that is in memory? (I call
this 'sync level data' for lack of a better term)
B) Is there any way of protecting this data? Or at
least tracking (journaling) what data is there so that when
the switchover to the new machine occurs, my customer can
be sure of the integrity of their data? (ideally this would
occur independent of the application)
C) What happens to this 'sync level data' in a disk
mirror take over in the event of a disk crash?
3) Do all of the disk drives need to be attached to
single serial subsystem adapter? In other words, if I had
2 serial subsystem adapters in the 980, each attached to 8
drives... could the 580 have a single subsystem adapter
which would attach to all 16 drives?
4) Is there an optimum serial cable length for both the
serial optical channel as well as the serial disks? (the
configurator provides many lengths so could you please
explain and comment on this... then make recommendations)
5) Does the 980 support AIX windows across a network?
(attaching an Xterminal or XONDOS based pc to the ethernet
adapter for example)
6) What are the cable choices for the hacmp serial link
(heartbeat)? Optical serial, copper (8-wire rs232) serial,
network....?
7) Can the 580 come with a serial disk select like the
980 has so that ALL disk on the system is serial disk?
8) In the event we do go with terminal servers rather
than ascii controllers....
A) What are the practical telnet limitations on an
ethernet card (before congestion or whatever else
becomes a problem)
B) does rlogin support transfer of emulation?
Their terminals are addsvp and are limited to tvi925
emulation only˘ (so telnet being limited to vt100
or vt220 is almost out of the question...maybe
rlogin provides a workaround so that risc/220's
could be used as terminal servers)
C) What are the practical rlogin limitations on an
ethernet card (before congestion or whatever else
becomes a problem)
---------- ---------- ---------- --------- ---------- ----------
A: I will answer your questions in the order that you asked them.
1. a) Netware/6000 can run with HACMP/6000, however, whenever
HACMP does an adapter swap, IP-address takeover, or when a failed
machine is brought back online, NETWARE/6000 will get clobbered and
have to be restarted.
b) Yes, you can attach the serial disks to both the 980 and
the 580 RISC System/6000 machines.
c) Both of these methods will work fine. However, if you get
a terminal server, i.e. 220 or 3XX machine, and have this server all
of the incoming terminals, it will be able to automatically switch to
the backup server. This is because the terminal server runs a HACMP
program called clinfo. This program keeps track of the cluster state
and does certain tasks accordingly. These tasks are manually
made by the user.
d) There are a couple of issues that you may want to consider.
The Serial Optical Channel Converter (SOCC) is not required by HACMP
and is not usually recommended unless the SOCC network is going to be
used for other private stuff between the two systems. To use SOCC just
for the heartbeat is considered to be overkill. HACMP/6000 now
requires that there is a raw RS-232 line connected between the two
machines. This RS-232 line is used to guard against TCP/IP failure on
one of the machines causing a failover to automatically occur.
2. a) When a failover occurs, all of the data in memory that has
not been synced will be lost.
b) Since there is no way to determine when a failure is going to
occur, there is not any way to completely record what is in memory. A
way to improve recovery of this data is to have the system sync more.
Or you can also be sure that a sync is done after every transaction.
This however will not completely guard against memory data loss.
c) I the event of a disk crash and having a mirror takeover, the
data in memory will still be there. Since the system will still be up
and active. Mirroring takeover is done on an LVM level and will not
affect the data in memory. When the sync is done, the data will be
written to the good copies of the logical volume.
3. No, all of the disk drives do not need to be connected to a
single serial subsystem adapter. As long as all 16 drives will reside
within 4 drawers, there is not any problems with putting them all on
a single adapter.
4. No, there is not an optimum serial cable for either the SOCC or
the serial link cables. The SOCC will be dependent on how far apart
the 980 is from the 580. Probably, since there will be a 7202 between
them, you will need to have at least a 10m cable. But again, this will
be dependent on how far apart you systems are. For the serial link
cables, it will be dependent on whether you will be going from a system
to a drawer within the system rack, or if you are going between a
system and a drawer in an external rack. Going from the 980 to 9333's
within it own rack, you will only need a 3m cable. But going from the
980 to the 9333's within the 7202 will require a 10m cable.
5. Yes, you can support AIX windows across the network. The only
restriction with the 980 is that it does not support an HFT (graphic
adapter).
6. The heartbeat will go across all defined networks. For the raw
RS-232 line you will need to have the following:
a. Printer/Terminal Interposer
b. 10-pin to 25-pin Jumper
c. 3m EIA-232 cable (25-pin on both ends)
d. Serial Gender Changer
7. No, according to the Salesmanual, the serial disk select
is not an option for the 580 RISC System/6000.
8. a) The practical telnet limitation will be dependent on what you
are running across that telnet session. What I suggest is that you
start up one average telnet session that you will be using. Then use
netstat to determine how much of the ethernet that one telnet session
is taking up. Saturation usually occurs at 70-80 percent of the
bandwidth. So use the netstat information to determine what the point
of congestion will be.
b) We have not heard of any terminal limitations with either
telnet or rlogin. When either of these are started up, there should
be an arbitration for the terminal type and if the term type is known
to the system that you are logging in to, it should be set to that
terminal. The tvi925 emulation is know to the AIX 3.2 machines and
should not be a problem. This emulation is located in the televideo.ti
file in the /lib/terminfo directory.
c) Determining this is the same as with telnet. It is hard to
know what the number of rlogin sessions there can be before saturation
of the network.
---------- ---------- ---------- --------- ---------- ----------
This item was created from library item Q612901 BWQRT
Additional search words:
BWQRT CONFIGURATIO HACMP IX JAN93 OP PROPOSED QUESTION RISCOSO
RISCSYSTEM SOFTWARE SYS
WWQA: ITEM: RTA000027792 ITEM: RTA000027792
Dated: 11/1996 Category: RISCOSO
This HTML file was generated 99/06/24~12:43:10
Comments or suggestions?
Contact us