[Info-vax] HP "cloud" disaster-recovery 'continuity solution' (... 90 minutes)

[Stephen Hoffman]

On 2012-08-16 11:36:19 +0000, MG said:

> The latest video of HP, on YouTube:
> 
> 	<http://www.youtube.com/watch?v=KzfMBdVTT4o>
> 
> 
> In the video description, it reads:
> 
> 	"HP Discover 2012: Session showcases recovery time
> 	frames following a disaster. Learn how HP's cloud
> 	continuity solution can easily and inexpensively
> 	recover your IT environment with only minutes of
> 	data loss and approximately 90 minutes of downtime."
> 
> 
> So, ~90 minutes of downtime... "[...] very fast recovery
> times"?  (A direct quote from the video.)


That's the HP "Disaster Recovery as a Service" and "HP Enterprise 
Services Cloud Continuity" video.

As with most things in computing, uptime and business-continuity 
requirements exist across a spectrum of requirements; from rolling in 
tapes and restarting development servers, to full-on replication and 
continuous uptime.    Different sites have different requirements.

It appears you believe the answer is always OpenVMS, and (for this 
case) DT clustering.  JF thinks the answer isn't VMS.  There's a whole 
lot of room between those two positions.  The costs and requirements of 
the range of available solutions and alternatives varies, too.

For this video and this example?

This certainly looks to be a very good deal for a site that can survive 
with their servers and applications offline for an hour or two; sites 
that don't want or don't need continuous service and its associated 
(large) costs.  That's a lot of sites.

For this approach, one hour nineteen minutes looks pretty good, and 
could well even be useful for those folks running OpenVMS in a data 
center and that don't want to deal with clustering and cluster-aware 
applications and the related costs.  (Yes, various folks here do know 
how to write cluster-aware and DT-aware software, but far from all of 
the OpenVMS applications around are actually written that way.  And 
fewer and fewer sites are writing their own software.)

The example configuration that HP cites in the video involves (mostly) 
SAP and an aggregate of 21 servers, comprised of 11 physical and 10 
virtual hosts, reportedly mostly Microsoft Windows with some Linux 
boxes involved.

Data (and application state!) was replicated to the HP data center.  
(VMS can't do application-state backups, BTW; not without application 
assistance or without a database.  This limitation is tied to why 
online backups are potentially unstable/unrecoverable on OpenVMS, 
outside of databases, too.  See the mess that is BACKUP 
/IGNORE=INTERLOCK as a start.)

The HP recovery process involves spooling up virtual machines from a 
pool of virtual servers, or spooling up virtual servers on dedicated, 
reserved server hardware, and rolling in the customer data, and dealing 
with the DNS TTL caching, and a whole pile of network routing with the 
telco(s) involved, restarting AD and DNS, and the rest of the 
application stack.

Yes, OpenVMS and Linux and BSD servers (and likely many others, these 
days) can be configured to maintain DR certainly recover vastly faster 
than this approach, too.  But at a (big) price, both in hardware, big 
data links for shadowing or other data replication, and the staff and 
skills to run that replication.

Or that you're hosting applications and replicating your data across 
(for instance) Amazon availability zones.

Testing factors into any of these, but the HP offering has HP helping 
with testing on the redundant servers.

Put another way, I'd be willing to bet that this service is a good and 
cost-effective option for Microsoft Windows Server configurations in 
particular - and that's a very big market - and that this service may 
well be a good (and cost-effective) option for those not inclined (or 
funded) to run a full DR configuration on Linux/BSD/OpenVMS/etc servers.

I don't know if this HP service would be available for OpenVMS, though. 
 And I don't know how HP would replicate OpenVMS application state 
here, outside of database-level backups, or outside of working with the 
specific package or with the software vendor.  And with what was 
identified as the HP data center hardware, the use of those x86-64 VM 
servers means that OpenVMS would have to be emulated, with all the 
baggage that brings, or would require a special pool of expensive 
servers with fewer users to spread the costs.  (Read: more expensive 
than x86-64 platforms.)

HP is listing this service as 20% to 50% cheaper than replication or 
repurposing a development/test configuration, and with no capital 
equipment; that's savings through staff and gear sharing at the remote 
data center, and a different bucket for taxes and related 
considerations.

And this means not hosing the development servers during the recovery, too.

This plan is also very close to outsourcing the whole data center.  To 
an HP data center, in this case.  HP is almost certainly aware of that 
potential, too.

This service looks like a pretty good deal.  Somebody would have to 
identify the number of "nines" for your uptime requirements, run the 
finances for your businesses, and run some tests to see how good a deal 
this really is, though.

> Linux and Windows are clearly the way forward, there's no
> place for that strange contraption called VMS...

The "contraption" called OpenVMS is competing in a much bigger market 
of products and services.  A market that wants and that needs various 
applications, that demands GUI interfaces and tools, that wants and 
needs trained or experienced staff, that needs 
backup/management/whatever client software, that needs integration with 
AD/LDAP/Kerberos/NTLM/whatever services, that requires competitive 
software and hardware features and capabilities, that particularly 
means/needs/requires cost-effective solutions, etc.  At a price that 
the customer can afford.  If the hardware and software platform does 
not meet these or if it's too expense to succeed in a market, then the 
"contraption" will be a specialized purchase.  In the case of OpenVMS, 
that usually means existing applications and existing clusters.




-- 
Pure Personal Opinion | HoffmanLabs LLC