Terry L. Inzauro wrote:
> Quanah Gibson-Mount wrote:
>>
>> --On Thursday, October 12, 2006 9:04 PM +0200 Michael Ströder
>> <[EMAIL PROTECTED]> wrote:
>>
>>> HI!
>>>
>>> I support Giovanni's statement that fail-over should preferrably be
>>> implemented in LDAP clients. Because the client really knows what to do
>>> if all fails. In practice this means to really teach the developers hard!
>> Sadly, Stanford has plenty of proprietary software that accesses our
>> LDAP servers.
>>
>>> Regarding TCP-level load-balancers: Typically the servers are located in
>>> one data center. IMHO this works quite well for most read-only
>>> applications. If you need fail-over for write requests you either need
>>> multi-master replication (with lack of data integrity) or some sort of
>>> hot-standby server which completely takes over the service.
>>>
>>> Also with using proxy servers for balancing the requests you should
>>> carefully think about your network topology to avoid WAN traffic to
>>> other data centers.
>>>
>>> If you'd like to implement fail-over between different locations
>>> probably DNS gets involved. But not simply DNS round robin!
>>
>> Stanford uses a software load balancer to a pool of replica systems (in
>> our case, we have a single source system for all writes, so we don't
>> worry about applications writing to the directories).  The main issue
>> I've found with using the software load balancer is that Java caches the
>> returned IP, so you need to do some junk to get it to do retries if the
>> connection to the system it is talking to fails.
>>
>> The software we use lets you run external commands on each individual
>> system that help determine the weights, so it is more than simple DNS
>> round robin.  For example, in Stanford's case, we have it run a local
>> ldapsearch to verify slapd is running.  If it is successful, then it
>> uses the system load to determine the weight.  If it is unsuccessful, it
>> makes the weight extremely high, essentially dropping it from the pool.
>>
>> --Quanah
>>
>> -- 
>> Quanah Gibson-Mount
>> Principal Software Developer
>> ITS/Shared Application Services
>> Stanford University
>> GnuPG Public Key: http://www.stanford.edu/~quanah/pgp.html
>>
>> ---
>> You are currently subscribed to [email protected] as:
>> [EMAIL PROTECTED]
>> To unsubscribe send email to [EMAIL PROTECTED] with the word
>> UNSUBSCRIBE as the SUBJECT of the message.
> 
> 
> can i ask what you use for your software based load balancer  linux lvs or is 
> it a "stanford" in-house application.
> 
> 
> _Terry
> 
> 
> 
> 

oops.  please disregard my previous post.  it helps to read further down the 
thread ;)

happy friday the 13!


_Terry

---
You are currently subscribed to [email protected] as: [EMAIL PROTECTED]
To unsubscribe send email to [EMAIL PROTECTED] with the word UNSUBSCRIBE as the 
SUBJECT of the message.

Reply via email to