[Info-vax] Moonshot

[Stephen Hoffman]

On 2015-01-07 07:40:51 +0000, David Froble said:

> Stephen Hoffman wrote:
> 
>> On 2015-01-06 09:17:21 +0000, Matthew H McKenzie said:
> 
>> For those that need that.  But there's a divergence happening, here.  
>> Where the hardware and the cores are outstripping what many folks need. 
>> For those that like their own metal where they can touch it (TaaS™), 
>> the available servers can be massively more capable than necessary.  
>> More than a few of the available server cores are idle, in most places. 
>> Sure, there'll be any number of folks looking to consolidate onto 
>> denser boxes.  Over time.  Incrementally.
> 
> Now, it's well known that I don't get out much, so consider that.
> 
> The question that I have is what is the real demand for the huge 
> systems being discussed here?  Sure, there is google and Amazon, and 
> some others.  But it appears that some of these are going their own 
> way. Perhaps no business for a vendor to sell into.

What may have been missed in my phrasing in that reply is that Moonshot 
<http://hp.com/go/moonshot> _is_ aimed at many of the "smaller" folks.

The folks that need big boxes need really, really, really big boxes, 
such as the Sierra and Summit servers: 
<http://info.nvidianews.com/rs/nvidia/images/An%20Inside%20Look%20at%20Summit%20and%20Sierra%20Supercomputers-3-1.pdf> 


Again, Moonshot is aimed at relatively smaller customers.   It's a 
dense system.   Up to 45 servers.    If you're consolidating a rack or 
two of servers in your existing data center, it's going to be of 
interest.   Alternatively, if you've decided to run, for instance, thin 
clients out at your user desktops, Moonshot can support the back-ends 
for your desktop displays.  With this thin-client configuration, you 
can swap the client or the cartridge if something goes wrong, and you 
can set aside a few boxes near the desktops and provisioned in the 
Moonshot, so that you don't have to rush the repairs.  Retrieve the 
user data from backups or from a SAN.  Various folks have suggested 
using Moonshot as a pool of web servers as well, and there are other 
approaches here such as database environments based on Cassandra or 
Aerospike.  (FWIW, Aerospike running on 50 hosts claims a million 
database writes per second, which isn't all that shabby.  I'd wager 
it'd be less than that on a Moonshot box, given the various obvious and 
probable differences in the Moonshot hardware from the Aerospike test 
configuration.)

It'd also be interesting to see whether a box-level interconnect in 
Moonshot could be used to allow some form of VMS clustering in addition 
to network-based clustering, but that's not going to be of interest to 
us end-user folks for a few years, and there'll probably be a 
newer/better/faster Moonshot box by then, too.  Hopefully VMS has much 
better mass-deployment and provisioning and distributed LDAP 
capabilities by then, too, as installing 45 VMS systems at a time using 
only the supported and documented means would get pretty tedious.  But 
I digress.

Now HP Apollo <http://hp.com/go/apollo> is a rather bigger system than 
Moonshot, and I wouldn't expect HP to be selling those boxes in large 
volumes.  Apollo is not quite in the Sierra and Summit range, but it's 
up there.  The NREL Peregrine is not a small configuration.

> So, back to "real demand".  What percentage of computer users have any 
> need or use for such large systems?  None of the users I'm working 
> with.   Note initial disclaimer above.

DEC was offering AlphaServer DS10L, with configurations of ~42 to a 
rack.  Anybody that was cruising eBay a few years back would know that 
DEC sold a number of those racks, too.   Moonshot is that rack, 
distilled down into a ~4.3 unit box, and much more power- and space- 
and cooling-efficient.

> Are some of the advances advantageous to many users?  I'd thing so.  
> But as mentioned, some users can get by with a rather small 
> configuration. The advances in speed and storage capacity have resolved 
> most of their needs.

Yes.  That's one of the reasons I was mentioning a whole lot of the 
current VMS folks might (will?) be happy with a one- or two-socket 
x86-64 box in place of their existing OpenVMS server, back in that 
earlier gonzo-core-count thread.  For the folks that want a 
rack-equivalent of those one- or two-socket boxes, there's Moonshot.

> Getting back to VMS.  From my rather restricted perspective, if VMS was 
> supported, supported current tools, supported current HW, and wasn't 
> missing anything rather needful, I think it would satisfy the needs of 
> most users, and would be successful.  Even with some of the warts.

Many of the current VMS users, yes.  Getting new users to consider and 
then perform new deployments is probably going to require some 
wart-scraping, and some feature upgrades.


-- 
Pure Personal Opinion | HoffmanLabs LLC