Re: Larry Ellison comments on Microsoft's benchmark

 Ivana - you haven't quoted the bit of my question that is most important. You are talking about a system where the  loss of one "node" means the entire system is then useless/broken. However many "business" systems can be achieved  with a partitioning of information across the (for instance) 10 nodes. Then if any one server/database/disk subsystem etc.  fails then you only lose 10% of the data. You can still get at the other 90% quite happily. 10% of your  customers can't get at that info until you recover that node - but the other 90% are unaffected.  

 The longest down-times I've seen for my customers have been caused by (a) database corruption, (b) some  idiot developer dropping a production database/table by mistake and (c) complete disk subsystem outage. OPS  doesn't protect you against any of these - a partioned system does to an extent. On the two occasions I've seen an actual "server" fail with a dodgy motherboard or memory the systems have been up again real quick with replacement parts or a spare server  just attached to the external raid and booted.  

 For SOME business requirements the OPS way makes most sense. For SOME business requirements the multiple-node  clustering makes more sense. For SOME an HA solution like Marathon/Stratus might be the right way. The only  thing that everyone should agree on is that there is no *RIGHT* way that is always best all the time for  every business requirement.  

 > > How can a single disk-subsystem be 12 times more reliable than
> > 12 disk subsystems? A logical or physical corruption here is
> > going to affect all OPS nodes.
>
> Yes, a corrupted disk will take affect all OPS nodes. However
> this is better than having 12 disks, and having your system go
> down if any *one* of those 12 disks is corrupted.
>
> If you have 12 disks as opposed to 1 then the probability of
> at least one disk going down is 12 times higer. So your system
> is 12 times less reliable.
>
> If you set the disk aside for a moment and look only at the
> reliability of the machines (and OS, especially in Windows land),
> then Oracle Parallel Server with the same 12 machines is INFINITELY
> more reliable than a DB/2 or MS SQL on 12 machines. Why? Because
> in Oracle Parallel Server, if one machine goes down the system
> as a whole is unaffected (except for lower performance.) Thus
> unless all the machines go down at the same time (very unlikely)
> your system is up. In DB/2 or MS SQL, if AT LEAST one machine
> goes down the system as a whole is down. If you have 12 machines
> then the probability of at least one machine going down is
> 12 times higher, so your system as a whole is 12 times less
> reliable compared to a system with only one machine. And if you
> compare to Oracle Parallel Server running on the same 12 machines,
> DB/2 and MS SQL are INFINITELY less reliable.

 Neil Pike MVP/MCSE. Protech Computing Ltd  (Please reply only to newsgroups)
 SQL FAQ (484 entries) see
 forumsb.compuserve.com/gvforums/UK/default.asp?SRV=MSDevApps (faqrtf.zip - L7 - SQL Public)

 or www.ntfaq.com/sql.html
 or www.sql-server.co.uk
 or www.mssqlserver.com/faq