Very Computer

Hi,

I was wondering what is the lowest cost server for high availability
purpose.  Basically, the server, with multiple net interfaces, runs
typical sys processes and a licensing server process.  Hopefully, it
will meet 99.9% uptime, annually.  That is including downtime for
maintance, if ever needed. :)  The platform is Solaris 2.6 or above.

Thanks in advance.

--
Image is not everything. Excellent image is.

Sent via Deja.com http://www.deja.com/
Before you buy.

Quote:> Hi,

> I was wondering what is the lowest cost server for high availability
> purpose.  Basically, the server, with multiple net interfaces, runs
> typical sys processes and a licensing server process.  Hopefully, it
> will meet 99.9% uptime, annually.  That is including downtime for
> maintance, if ever needed. :)  The platform is Solaris 2.6 or above.

99.9% uptime isn't high availability. That's almost 9 hours a year when
nothing works, or 2 hours if you only consider 9-5 during the week (but
then you can do maintenance outside working hours).

Pretty well any Sun around will get you that far, provided you have
mirrored system disks and mirrored or raid-5 data disks. You'll
probably want to have a standby system just in case the whole box goes
AWOL, and make sure you have a UPS. Your requirements of the box don't
sound overly taxing - probably best to keep it as simple as possible
(our bigger servers have failures more often, simply because they have
more components, but even our worst machines are better than 99.9%,
although in one case it's only just...). And the simplicity includes
keeping people away from the box who may have an urge to fiddle with
it.

--
-Peter Tribble
HGMP Computing Services
http://www.hgmp.mrc.ac.uk/~ptribble/

: Hi,

: I was wondering what is the lowest cost server for high availability
: purpose.  Basically, the server, with multiple net interfaces, runs
: typical sys processes and a licensing server process.  Hopefully, it
: will meet 99.9% uptime, annually.  That is including downtime for
: maintance, if ever needed. :)  The platform is Solaris 2.6 or above.

The problem is that for HA you are going to need at least 2 servers
with identical configurations using shared disk. Look into either
E250 or E450 servers and A1000 or A5000 shared storage arrays running
Sun Enterprise Cluster Management software.

Hardware-wise you could get by with less (like a couple of the Ultra
workstations, Ultra 10's and such), but you would probably outgrow them
very quickly.

fpsm
--
| Fredrich P. Maney                                                   |

|          Fight Spam! Join CAUCE! == http://www.cauce.org/           |
|                Outlaw Junk Email! Support HR1748.                   |
|          "Sometimes, fear has a good and useful purpose."           |

Say if I need two or more servers for the licensing server process,
either clustering config or failback config.  How would the hostid
going to be assigned?

Sent via Deja.com http://www.deja.com/
Before you buy.

: Say if I need two or more servers for the licensing server process,
: either clustering config or failback config.  How would the hostid
: going to be assigned?

That question didn't really make any sense. However, I am going to try
to answer it.

In a Sun Cluster environment (regardless of type), you install the
operating system and configure the machines as individual machines first.
Then you install the Cluster software and then the Volume Management
Software, then you create the virtual hosts and processes.

So, you would have (at minimum): 2 Servers, 1 Adminin Workstation, 1
terminal server, and one external disk cabinet. You would have to have
a minimum of 5 ip addresses. Your host table on each machine would look
something like this:

####
127.0.0.1 localhost
1.2.3.1   aws       # Admin Workstation
1.2.3.2   tc        # Terminal Concentrator
1.2.3.3   node0     # node0
1.2.3.4   node1     # node1
#1.2.3.5   node3     # node3
#1.2.3.6   node4     # node4
####

However, you really should have duplicate cards (2 qfe cards work great)
in each server to allow failover between the cards before having to
failover to the other machine. In other words, each server would have at
minimum 2 physical interfaces (in addition to the private cluster network
interfaces) for every virtual host/process.

fpsm
--
| Fredrich P. Maney                                                   |

|          Fight Spam! Join CAUCE! == http://www.cauce.org/           |
|                Outlaw Junk Email! Support HR1748.                   |
|          "Sometimes, fear has a good and useful purpose."           |

Fredrich,

Thanks for the great post!  Could you comment on how to find out
failures on network card, CPU, disk, or board?  Is there auto detect on
the clustering software to alert the admin?  Thanks again.

Sent via Deja.com http://www.deja.com/
Before you buy.

: Fredrich,

: Thanks for the great post!  Could you comment on how to find out
: failures on network card, CPU, disk, or board?  Is there auto detect on
: the clustering software to alert the admin?  Thanks again.

Your best bet here is to go to the Sun Cluster class offered by Sun
Educational Services or to at least buy the SunEd books on Clustering.
I *very* strongly reccommend tht you take the class. I don't have my
book handy right now, and it's been a few months since I messed with
this stuff.

However, the Sun Cluster software comes with specific monitoring software
that watches the various network interfaces (using PNM - Public Network
Management and NAFO groups) both locally and remotely (the nodes monitor
each other). There are specific tools to monitor Sun supplied processes
(like Oracle, DB2 and NFS) and you can write your own fairly easily
for your own homegrown tools.

Also, the Sun Cluster software only monitors the systems in order to
handle failover. If you want to monitor the machines health and be
alerted to any problems, you could add any of several products (I would
suggest BigBrother (www.bb4.com)) to the mix to monitor the various
hardware and the various log files and processes (including those for
the cluster software). This software would then alert you, either
through email, a web interface or by paging you.

fpsm
--
| Fredrich P. Maney                                                   |

|          Fight Spam! Join CAUCE! == http://www.cauce.org/           |
|                Outlaw Junk Email! Support HR1748.                   |
|          "Sometimes, fear has a good and useful purpose."           |

The hostid is on the prom. Be prepared to pay for an additional
license for the second server.

--
-Andy M
http://synecdoche.net/~andy

Sent via Deja.com http://www.deja.com/
Before you buy.

:> Say if I need two or more servers for the licensing server process,
:> either clustering config or failback config.  How would the hostid
:> going to be assigned?

: The hostid is on the prom. Be prepared to pay for an additional
: license for the second server.

However, in clustering, the logical host is addressed by the hostname/IP
that is assigned to the logical host, not by the hostid of the physical
server it currently resides on.

fpsm
--
| Fredrich P. Maney                                                   |

|          Fight Spam! Join CAUCE! == http://www.cauce.org/           |
|                Outlaw Junk Email! Support HR1748.                   |
|          "Sometimes, fear has a good and useful purpose."           |