Feeds

back to article HP's 'historic' Project Moonshot servers aim at hyperscale future

HP didn't invent the rack-server business, but Compaq – the company it acquired more than a decade ago – did. HP can't buy its way into the next system era, which is why it is trying to create that era itself with its second-generation Project Moonshot servers. The initial "Redstone" Moonshot boxes were development machines …

COMMENTS

This topic is closed for new posts.
Silver badge
Devil

This can become good.

But ... where is the RAM on that Gemini server node?

And ... Itanium must have been airbrushed out of the future.

1
2
Silver badge
FAIL

Re: Destroyed All Braincells Re: This can become good.

"....Itanium must have been airbrushed out of the future." Don't be silly, that would be like using a Ferrari as a milk float. These are aimed at small workloads like hosted webservers and the like, not enterprise big iron. Please do try and learn the difference.

1
0

The Meg and Dave Show

Too little, too late. Too little, as in 8GBytes, and too late, as in it's 2013.

0
1

Captain Server Pants

They make it hard to compare the specs on this thing. It seems like a 4.3U server contains 3 rows of 15 cartridges with 4 nodes each. Each node is 64-bit Atom 2.0 GHz and has 8Gb Ram. So 180 nodes with 360 threads and (8x45) 360 Gb of Ram.

Why the odd rack size? That's ok if they fill up the entire rack, like with a Superdome.

If you compare this machine to an Oracle T5-4 it's pretty clear HP is out to lunch. Will you be able to run licensed code on this box? Which? What about HP-UX?

0
1
Silver badge
Boffin

Re: Captain Server Pants

"....Why the odd rack size?...." Agreed, very good question. Why not just make it 5U tall and put a little more airpsace in the chassis, you'd still get just as many in a standard 42U rack? Or is it a weight/heat thing, like the blades, where if you put too many in a standard 42U rack you go beyond the weight limit (according to hp, four fully-stacked C7000 blade chassis are too much weight in a standard 42U rack)?

And why a 500GB SATA drive, surely you would want an SSD to lower power consumption?

"....If you compare this machine to an Oracle T5-4 it's pretty clear HP is out to lunch....." Well, if you think that they are both aimed at the same target - weiner thread webservers - and that the hp option will run bogstandard x64 apps that the Ts can't, then suddenly it makes more sense.

"....What about HP-UX?" Webserver market - not much call for an enterprise UNIX at the low end when Linux does it cheaper.

I wonder if they'll make a blade that slots into the C7000 chassis so they can be mixed with Xeons and Itaniums and also take advantage of Flex Fabric.

0
0

Re: Captain Server Pants

Ok so you say:

+ "... will run bogstandard x64 apps"

+ "...not much call for enterprise Unix at the low end when Linux does it cheaper."

+ "...I wonder if they'll make a blade..."

Help me understand the hyperbole from HP management. Why are people going to be looking back at this announcement as huge news? I'm longtime database developer primarily on AIX and HP-UX platforms. This seems like another low end box where HP dumps you off on the door step of a Linux vendor or Microsoft. Are we supposed to forget about HP-UX? Idk it just seems odd to hear a major server announcement with major hoopla and leave out or forget mention the rest of HP servers.

You say this is a low end box and so it appears. One vendor might call their box a low end machine and charge a high price, another might call it a high end box and sell to for a low price. If you forget about vendor offered characterizations, the T box has way more of everything in the same (or roughly the same) rack space.

0
2
Anonymous Coward

Re: Captain Server Pants

That 0.7U does make a difference - it's the difference between 8 and 9 of these in a standard 42U, or I'm sure it said somewhere about 10 in a 43U rack.

0
0
Anonymous Coward

Re: Captain Server Pants

I think you're missing the point, this isn't targetted at heavy lifting, this is targetted at scale out workloads. It's not a replacement for the current Proliants, it targets a different market altogether.

2
0

Re: Captain Server Pants

"I think you're missing the point, this isn't targetted at heavy lifting, this is targetted at scale out workloads. It's not a replacement for the current Proliants, it targets a different market altogether."

Am I the one missing the point? You listen to what vendor says and make conclusion. I look at what box can do in data center and then decide. T series is for scale out also but flexible and probably only for current Oracle customers. How does new box compare to 4 DL100's with 2x low voltage 8 core Xeons? That's 64 cores and 128 full power threads in 4U. I'm having a hard time configure situation in which new boxes make sense. Forget out data layer. That's my opinion with much experience. HP doesn't want to talk about data layer. My opinion based on article, if virtualization penalty is high on Atom and machine is used as front end or file utility then machine makes little to no sense when you consider alternatives. If machine exists in mixed use environment then it's useless. If make custom box for oneself, then immediately discard all 45 cartridge casings.

0
1
Silver badge
Boffin

Re: Captain Server Pants

"....virtualization penalty is high on Atom...." Your using the wrong implementation model. With multicore Xeon and Opteron, we've got used to the idea of carving up boxes with virtualization software because we had so many powerful cores in even a 2-socket server. Instead, I would suspect that the model with these systems will be that of cobbling individual Atom server blocks in grid-like cluster instances. Granularity is stuck at the Atom CPU level, but management is easier, and you increase or decreas the power of an instance by adding or removing nodes from the cluster. The result is also more resilient seeing as the load is split over several disparate nodes, especially if you can build your instance across two or more chassis. If you imagine using a 2-socket server with hex-core Xeons for a VM that needs 8 cores, if you lose a CPU you could lose most of your processing capability. With this type of mini-server board, if you lose a board you lose two cores, you simply carry on at 75% processing power whilst you switch in another card to bring the cluster back up to 8 cores (you might even have an automated solution that adds in a spare card to the cluster the minute one fails).

1
0
Silver badge

Progress chart

I quite like the idea that the x86 SystemPro represented an advance on the 840.

1
0
WTF?

taller racks then?

"8 to 10 million new servers in the next three years. Using convention rack or blade servers, those servers would take up around 200 football fields – about the 13.5-mile length of Manhattan"

Gotta laugh at this crappy unit of measurement. Why would a server be on a football field?

So anyway the answer is obvious, given that the analogy is Manhattan.... 100ft high racks!

0
0
Silver badge

With 2 disks...

...they could also have snatched the root-server market.

0
0
Anonymous Coward

Re: With 2 disks...

I think there is a cartridge which houses 2 drives. And when you throw memristor into this, it will fly. HP are already running part of their web site on this technology, using the same power as 2 lightbulbs!

1
0
Silver badge
Meh

Re: AC Re: With 2 disks...

".....HP are already running part of their web site on this technology....." Well, two thumbs up for eating their own dog food, but as someone that regularly suffers the hp public websites I'd have to say their web designers produce what comes out of the other end of the dog.

0
0
Silver badge

Re: With 2 disks...

"HP are already running part of their web site on this technology"

I'll believe it's potentially viable when they start shipping eval units. Until then it's a science project, just like bubble memory and other friends.

0
0
Anonymous Coward

Re: With 2 disks...

Eval units? HP have had Moonshot eval units out for months, it's a commercially available product as of now!

2
0
Alien

Power

"And those data centers will cost on the order of $10bn to $20bn, she said, and would require approximately ten power plants to juice them up – about the equivalent of two million American households."

Maybe Google, Facebook, and Amazon, should be investing in WAMSR reactors?

http://transatomicpower.com/products.php

0
0
Meh

HP RIP

Anyone can make these. What makes it viable, both for HP and the customer, is the management layer, just like OpenView and Insight Manager made ProLiant and HPUX viable at scale in the days of steam power. (To be fair I would add that those boxes were reliable, which was also nice.) I'm not getting any software love from the Moonshot site, or from the article. What's the HP sauce in the firmware or the rest of the stack? Do I have to pay for that? Or is it just OpenStack? Which would mean: anyone can make these.

Btw including local storage lends a certain retro chic, but it's an odd way to save power. Tell me it's optional.

2
0
This topic is closed for new posts.