Re: clustering and high availability?

From: Daniel Morgan <damorgan_at_x.washington.edu>
Date: Thu, 10 Jun 2004 17:39:58 -0700
Message-ID: <1086914413.599812@yasure>

Michael Austin wrote:

> Daniel Morgan wrote:
>

>> Mike wrote:
>>
>>> We're starting a project at work moving VSAM to RDBMS. The choice
>>> is between DB2 and Oracle. It seems like the Oracle RAC is a better
>>> cluster choice with it's share everything rather than the DB2
>>> share nothing. Please post some opinions on this and/or other
>>> points of difference/intereste between the two DBMS.
>>>
>>> Mike
>>
>>
>>
>> Before you make this decision you need to test your application
>> in a RAC environment and see how the memory interconnect works.
>>
>> Assuming it is a well written scalable application consider the
>> following:
>>
>> Shared Everything:
>> The more nodes I add the mean time between failures goes up
>>
>> Shared Nothing:
>> The more nodes I add the mean time between failures goes down
>>
>> Shared nothing makes the problem worse ... not better
>>
>> Shared Everything:
>> Change the number of nodes and no change need be made to the
>> database.
>>
>> Shared Nothing:
>> Change the number of nodes and bring the server down while you
>> re-federate the data.
>>
>> DB2 is not in the ballpark unless running on OS/390 where it
>> is, in fact, shared everything. If shared nothing was better
>> you'd think IBM would have used it on OS/390 too: They didn't.
>> But who can afford to cluster mainframes?

>
>
> According to technical sales reps, RAC in a clustered environment only
> worked as intended - with 100% database availability across a cluster -
> on OpenVMS and Tru64 (5.1+)... say what you will about dinasaurs, but
> this is a technology that has been around for 20+ years now and no one
> has been able to duplicate it.
>
> It is still the only platform(s) that have true direct concurrent disk
> access 100% of the time(none of this NFS or active-passive crap). I
> have seen what happens to filesystems where more than one node tried to
> access a logical disk volume via a SAN or direct SCSI interconnect on
> those "other" operating systems and it ain't pretty...
>
> And if you want REAL clusterability you can still get Oracle Rdb
> (formerly DEC Rdb) for OpenVMS - why do you think one of the niche
> markets for this database and OS is stock market trading??? Because
> they don't want it to go down.
>
> Michal Austin.
> OpenVMS biggot :)
> but I can still unix and windows with the best of them...

Anytime you want to come to my lab I will be happy to give you four hours to try to bring down an 8 node cluster with RedHat Linux and a NetApp F810 Filerhead NFS mounted.

No one's done it yet.

-- 
Daniel Morgan
http://www.outreach.washington.edu/ext/certificates/oad/oad_crs.asp
http://www.outreach.washington.edu/ext/certificates/aoa/aoa_crs.asp
damorgan_at_x.washington.edu
(replace 'x' with a 'u' to reply)

Received on Thu Jun 10 2004 - 19:39:58 CDT