Welcome to our community

Dear Friends,

how can i utilize the memory channel in my trucluster V5.1B, two nodes GS80?

what utility can i use to know the performance of these memory channels?

Best Regards,

amr

Hello Amr,
You can use memory channel to create a TruCluster and use your GS80s. You need to use Memory Channel 2 hardware, and if you are creating a dual rail MC cluster you must set the MC window size to 128MB on both adapters if they are installed in the same PCI bus segment, otherwise 512MB setting will be OK. Measuring performance is a difficult one, I will look into it
Regards

Hi Paul,

thank you for your reply, can you tel me how to set the size on my memory channel?, as it is already came with 128MB as default, is it possible to increaase it to 512MB?

to let you in the picture with me, my site is making tunning in our system, and we found that the communication speed between the two nodes (Two GS80 with one memory channel in each) is slow (not enough for performance), so how can i over come this problem, please?

Best Regards,

Amr

Hi,

I haven't used TruCluster in earnest for over a year now but I remember the issue of a seeming lack of tools to show memory channel performance. I think in 5.1B and later the cfsstat command was improved to show more details of the use of the memory channel. The cfsstat command displays statistical information about not only the cluster file system (CFS) but also internode communications services (ICS).

You may already be familiar with this but if not you may try to check the icsstat, icschan and icssvc statistics displayed with this command.

regards

Simon

Hi Amr,
The mc jumper J3 needs moving from pins 1-2 (128M) to 2-3 (512M. You also should check the kernel attribute /sbin/sysconfig -q rm rm_use_512=1. On one cluster member it needs to be =1 to enable the 512M window size, the configuration change is propagated to the other cluster member systems by entering the following command:
/sbin/sysconfig -r rm rm_use_512=1.
Hope this helps

Hi Paul,

thank you for your care, i need to confirm somethings related to my MC changes, firstly is these changes in MC adapter and kernel parameter impact my system anywhere?

and i need to know about how to configure dual rail MC, is it required two MC adapter in each member?

amr

Hi Amr,
.
The changes to the kernel only affect the MC, is really only performance
issue. The dual rail MC setup requires that you install two MC adapters in
each machine, best performance is gained by installing the MC adapters in
separate busses. The MC adapter is a high throughput device despite being
only 32 bit interface, and should not be mixed with other devices except
another MC adapter or copper/fiber converter. You don't say if you have a
MC-Hub, so I will give instructions for both Standard hub and virtual hub
configuration. And I assume you were not running a LAN based cluster before.

With power off.
Virtual Hub, that is hub-less, requires you to set the jumpers to V-HUB-0
pins 2-3 on the two adapters in one system and V-HUB-1 pin 3 (jumper parked
on one pin) on the two adapters in the other system. Now power up the two
systems and at the P00>>> prompt run the mc_diag utility on both systems,
this runs an internal MC adapter diagnostic on both adapters, it is not
infallible. Now execute show config to find the MC adapter mca0, take the MC
link cable, the black one P/N BN39B-10 cable and connect mca0 on one system
and connect it to the other systems mca0, execute mc_cable on both systems,
you should see node 0 and 1 reporting in from the mca0 device, if it reports
that the report is from mcb0, you have the cables crossed. It is of to
connect cables with power on because the signals are LVDC. You enter ^C to
exit mc_cable, if you look at the two LEDs on the MC adapters they should be
both green whilst mc_cable is running and amber when mc_cable is stopped.
Assuming you have both ends connected you can now proceed to connect the
other cable and run mc_cable again, check the results. You are now ready to
form a cluster. No further configuration is necessary, both MC adapters
operate logically as one, although only one will be active the other will be
standby, you cannot tell which.

Standard HUB. The procedure is similar to VHUB, except that the Node ID is
determined by the HUB, this is achieved by setting the jumper to pins 1-2 on
all adapters. In a dual rail hub all the mca0 cables connect to the first
hub, it doesn't matter which link card you choose as long as you use the
same card (therefore node ID) on the mcb0 hub, the mcb0 path/cables connect
to the other hub, failure to comply with this will cause major problems.
Thus it is important to test using mc_cable all the mca0 paths first, then
all the mcb0 paths with the mca0 hub powered off before testing the mcb0
paths to avoid confusion. If all the links are checked out then you are
ready to form a MC cluster. You should also see that both LEDs on the MC
adapters and on the link cards in the HUB light green while mc_cable is
running and revert to amber when not.

When you form your cluster you will see in the boot messages confirmation of
your setup. You may choose to test the failover works, to do this select
MCA0 hub, this is likely to be the active path, and reset it or power it
off. The cluster should switch over to use the other path. We are never sure
if the cluster will switch back to mca0 when it is "repaired" so it is a
good idea to reboot the cluster after this exercise to ensure we revert to
the correct active path.

I hope this helps. Most of the setup listed above is found in the cluster
hardware config document on the V5-1B docs CD.
Regards

Hi Amr,
Further to your problem, with regard to the performance issue could be related to the patch level of your O/S. Check your patch level.
Regards

hi Paul,

thank you very much for your help, as i have just fix my problem now, by change the jumper as you adviced me, and it automatically change the kernel parameter rm_use_512=1.

i had already upgraded my system to patch kit 4, from along time, where i running tru64V5.1B as i mentioned to you before.

thanks again, and i hope to meet you in others topics

Hi Amr,
thanks for the reply, nice to know that you solved your problem, see you at Host some time.
Regards
Paul