Mapserver in a cluster

[Stephen Woodbridge]

cozzolongo wrote:
> 
> -----Original message----- From: Stephen Woodbridge
> woodbri at SWOODBRIDGE.COM Date: Fri,  3 Nov 2006 20:11:58 +0100 To:
> MAPSERVER-USERS at lists.umn.edu Subject: Re: [UMN_MAPSERVER-USERS]
> Mapserver in a cluster
> 
>> percy wrote:
>>> we are interested in this as well. we have a beowulf cluster, and
>>> would like to explore load-balancing in anticipation of increased
>>> demand for our geospatial data.
>>> 
>>> I wonder what Ed over at Topozone does :-)
>>> 
>>> Salvatore Cozzolongo wrote:
>>>> I would put mapserver cgi in a cluster (HA or HPC). The web
>>>> server should get up the cgi in this environment. According to
>>>> you, is it possible? what type of cluster do you try to use?
>>>> Best Regards to all -- Salvatore Cozzolongo
>>>> 
>> I can speak to how this was originally set up. I believe Ed has
>> made some changes. There are approximately 10 servers with the USGS
>> DOQQ data organized into one degree blocks and spread across the
>> servers in a somewhat random order. When a map request comes it it
>> is directed to the system that owns the one degree block at the
>> center of the map. All systems are cross mounted using a Samba file
>> system because of its ability to recover from connection failures,
>> which NFS can not do, so every system looks like it has the full
>> compliment of DOQQS.
>> 
>> I'll let Ed jump in to say more if he wants.
>> 
>> At Where2getit, we have a cluster with 8-9 servers behind a linux
>> load balancer. The load balancer monitors the client systems and
>> their load and dispatches requests to them respectively. These
>> systems are all severing vector data, and each is full compliment
>> of data and software. In fact each is a mirror of the master for
>> that cluster. We can pull any server from the load for maintenance
>> and the remainder will handle the load. If a server has a hardware
>> failure the load balancer will remove it from the load and
>> sysadmins are paged. If we need more capacity, we can load the
>> basic OS on a box, mirror it from the master and add it to the
>> load.
>> 
>> -Steve
>> 
> 
> Is this  a somewhat HA-type cluster with all data replicated on all
> nodes in a certain mode? do i understand well? May a  linux HA
> heartbeat system good for this  kind of cluster, no? Your reply is
> very interesting for my studies.  Very Very Very thanx...
> 
> 
> ------------------------------------------------------------------------
> 
> 
> 
> 000000000000000000000000 Date: Sat, 04 Nov 2006 17:17:09 +0100 From:
> cozzolongo <cozzolongo at planetek.it> To: "Stephen Woodbridge"
> <woodbri at SWOODBRIDGE.COM> Reply-To: cozzolongo at planetek.it Subject:
> Re: [UMN_MAPSERVER-USERS] Mapserver in a cluster Message-ID:
> <2eead1c61edecd4a60783be43b0c6f01 at planetek.it> X-Mailer: IceWarp Web
> Mail 5.3.0 X-Originating-IP: 213.203.172.165 MIME-Version: 1.0 
> Content-Type: text/plain; charset="none" Content-Transfer-Encoding:
> 7bit
> 
> 
> -----Original message----- From: Stephen Woodbridge
> woodbri at SWOODBRIDGE.COM Date: Fri,  3 Nov 2006 20:11:58 +0100 To:
> MAPSERVER-USERS at lists.umn.edu Subject: Re: [UMN_MAPSERVER-USERS]
> Mapserver in a cluster
> 
>> percy wrote:
>>> we are interested in this as well. we have a beowulf cluster, and
>>> would like to explore load-balancing in anticipation of increased
>>> demand for our geospatial data.
>>> 
>>> I wonder what Ed over at Topozone does :-)
>>> 
>>> Salvatore Cozzolongo wrote:
>>>> I would put mapserver cgi in a cluster (HA or HPC). The web
>>>> server should get up the cgi in this environment. According to
>>>> you, is it possible? what type of cluster do you try to use?
>>>> Best Regards to all -- Salvatore Cozzolongo
>>>> 
>> I can speak to how this was originally set up. I believe Ed has
>> made some changes. There are approximately 10 servers with the USGS
>> DOQQ data organized into one degree blocks and spread across the
>> servers in a somewhat random order. When a map request comes it it
>> is directed to the system that owns the one degree block at the
>> center of the map. All systems are cross mounted using a Samba file
>> system because of its ability to recover from connection failures,
>> which NFS can not do, so every system looks like it has the full
>> compliment of DOQQS.
>> 
>> I'll let Ed jump in to say more if he wants.
>> 
>> At Where2getit, we have a cluster with 8-9 servers behind a linux
>> load balancer. The load balancer monitors the client systems and
>> their load and dispatches requests to them respectively. These
>> systems are all severing vector data, and each is full compliment
>> of data and software. In fact each is a mirror of the master for
>> that cluster. We can pull any server from the load for maintenance
>> and the remainder will handle the load. If a server has a hardware
>> failure the load balancer will remove it from the load and
>> sysadmins are paged. If we need more capacity, we can load the
>> basic OS on a box, mirror it from the master and add it to the
>> load.
>> 
>> -Steve
>> 
> 
> Is this  an HA-type cluster? With all data replicated on all nodes in
> a certain mode? do i understand well? is A linux HA heartbeat system
> good for this  kind of cluster? Your reply was very interesting for
> my studies.  Very Very Very thanx for your reply..

We use a bunch of components similar to Ultramonkey. check out their 
site or google for  "LVS" or "Linux Virtual Server" would be a good 
starting place. Yes it is a HA-type cluster with a heart-beat.

-Steve