[Linux-cluster] Network-attached power switches for RHCS

Mohd Irwan Jamaluddin irwan at magnifix.com.my
Tue Oct 31 06:16:26 UTC 2006


On Mon, 2006-10-30 at 13:16 -0500, Lon Hohberger wrote:
> 1U APC switches are less expensive and have SNMP capabilities, but to
> feed NSPF power to a cluster, you would need at least two of them; the
> APC 9225 switches only have one power rail.  This means if you trip over
> the switch's power cable, all cluster nodes powered by the APC unit die.
> APC switches can also often be daisy chained together, but our agents do
> not generally support the configuration.  (I think the CVS version does
> on the most recent hardware, but probably not on the 9225...  Jim?)
> 
> Ex: single power supplies connected to typical APC devices:
> 
>               power1  power2
>                 |       |
>     server A   APC1    APC2   server B
>            +----1       1-----+
> 
> Ex: dual power supplies connected to typical APC devices:
> 
>               power1  
>                 |     
>     server A   APC1           server B
>     |      +----1             |      |
>     |           2-------------+      |
>     |                                |
>     |                  APC2          |
>     +-------------------1            |
>                         2------------+
>                         |
>                       power2
> 
> 
> 
> The WTI IPS800 switches above have two power rails with 4 ports each.
> The CE version is 208v; the non-CE is 110~120v, otherwise, they're the
> same.  They cost more per unit than the 9225.  You can control dual
> power supplies on separate rails, giving you NSPF as far as power-cords
> are concerned.  If you pull one of the power sources, the power switch
> is still accessible and fencing will still work, because the internal
> electronics can run off of either power rail.  (I'm not sure what the
> fault mode is if the electronics fail, though; I *think* it leaves the
> ports in their current states; contact WTI if you have questions about
> this).
> 
> Ex: dual power supplies connected to typical WTI devices:
> 
>               power1
>                 |
>     server A   IPS   server B
>            +----1    |
>                ===   |
>                 5----+
>                 |
>               power2
> 
> Ex: dual power supplies connected to typical WTI devices:
> 
>               power1
>                 |
>     server A   IPS   server B
>     |      +----1    |      |
>     |           2----+      |
>     |          ===          |
>     +-----------5           |
>                 6-----------+
>                 |
>               power2
> 
> >From a reliability standpoint, APC and WTI both make extremely reliable
> devices.  I've never had any switch from either vendor go bonkers on me.
> 
> Not to state the opinion of any company I may or may not be employed by
> or affiliated with, I personally generally prefer the WTI devices over
> APC devices because of:
> 
> (a) Design - I *totally* dig the dual power rail configuration.  It has
> higher power capacity per switch (30A, 15A per rail/4 ports), as well
> two power sources (note: it's *two* rails; not a single, redundant rail,
> even though the switch control electronics can run off of either rail)
> 
> (b) Firmware revisions on APC devices have broken fencing agents on more
> than one occasion. (Though, this isn't so much a problem with the newer
> APC SNMP fencing agents, but I don't like setting up SNMP...).
> 
> Jim Parsons (current fencing maintainer) has differing opinions on the
> matter; I believe he prefers APC units over WTI units.
> 
> -- Lon

Thanks Lon for all the superb informations.

Due to support availability, I prefer APC over WTI. It's just that APC
has office in Malaysia and provide support locally.

Now the problem is; I'm not be able to use the APC9225 model due to
output voltage issue. APC9225 uses 120V while we in Malaysia use
220-240V/50Hz. 

APC suggest me to use another model, APC7921[1]. Nevertheless APC7921 is
not included in Red Hat documentation as a preferred (supported???)
model. So, any idea on this issue? Is Red Hat going to support me even
though I'm using unrecommended model?

I have 2 servers to be clustered with a single service. Is it ok to use
1 network power switch, or is it recommended 2 power switches (1 power
switch per server)?

[1] http://www.apc.com/resource/include/techspec_index.cfm?
base_sku=AP7921
[2] http://www.redhat.com/docs/manuals/csgfs/browse/rh-cs-en/s1-
hardware-clustertable.html#TBL-HARDWARE-FENCEDEVS 

-- 
Regards,
+--------------------------------+
|       Mohd Irwan Jamaluddin    |
| ##    System Engineer,         |
| (o_   Magnifix Sdn. Bhd.       |
| //\   Tel: +603 42705073       |
| V_/_  Fax: +603 42701960       |
|       http://www.magnifix.com/ |      
+--------------------------------+
| "Every successful side needs   |              
| unsung heroes" - fcbayern.de   |      
+--------------------------------+




More information about the Linux-cluster mailing list