Scale UP or Scale OUT – The Storage side

Scale UP or Scale OUT – The Storage side
CINETICABLOG

Fabio
Rapposelli

18
marzo
2010

Hello there! If you are new here, you might want to subscribe to the RSS feed for updates on this topic.

I saw an interesting discussion between Duncan Epping and others in the Virtualization Twitterverse and I was wondering how the claims of upwards scalability in the storage world are effectively true or just plain fiction.

I take, as an example, Hitachi and its USP-V: they claim that the USP-V can address 247 Petabytes of storage behind it (virtualized and internal), and that the storage can sustain 4 MILLION IOPS!. That’s pretty impressive! (even if you read their SPC-1 Benchmark report and see that with their best config they can reach just 200.000 IOPS but that’s another story).

Well, it’s not so impressive actually

If you do some simple math you can clearly see that 247PB (maximum addressable space) divided by 4.000.000 IOPS (maximum IOPS achieved) give you a mere 0.01521 IOPS per GB, that’s really a NOT impressive.

So, even with the best configuration, best practices followed, best Tier 3 storage virtualized behind it you cannot squeeze the system to its maximum capabilities.

Considering that, I think that Monolithic storage is on a dead end railway, the Scale out approach is surely the way to go in the near future.

Tag: HDS, Scale Out, Scale Up, Storage, Virtualization

Articolo inviato il 18 marzo 2010 alle 12:35 nella categoria Storage, Virtualization.
Resta aggiornato abbonandoti al feed RSS 2.0. Vai al modulo dei commenti. Ping non attivo.

frapposelli

Calvin,
I'm looking forward to read your upcoming blog post, I'm eager to see your point of view on where the old monolithic storages still stand strong in terms of availability.
Calvin Zito

Thanks for clarifying this was your post Fabio. I disagree with your assessment that mid-range arrays can sustain the availablility levels of products like the XP. Disk Array. They can't. There's still a gap between that functionality. People have often questioned whether HP would continue to sell the XP - and the answer is we absolutely will because mid-range products can't deliver the same availability. I think you've given me an idea for a blog post on my blog.
frapposelli

Calvin,
It's not an Enrico's post, but mine.
And you obviously point the argument in the right direction, the Scale Up model is no more sustainable, HP did a great job acquiring LeftHand, they have a really good product based on standard components.
A next-generation scalable mid-range array right now can sustain the availability levels of the old-fashioned enterprise systems so that will not be an issue anymore.
It's Hitachi that seems to focus only on the monolithic approach, without considering the scalable approach, they hide behind their concept of "USP-V in front of everything" but unless they start to scale to a federation of them they will slowly go the way of the dodo.

Fabio
Calvin Zito

Enrico,
Disclaimer - I work for HP StorageWorks.

When looking at questions like the one you raise, I think you have to consider all the different considerations, not just one. People that use products like USP-V and the StorageWorks XP Disk Array use those products because they know once they turn the product on, they'll never turn it off. Not all data needs this kind of availability. That said, I think we all know that the high-end enterprise storage market has flattened out. I'm sure in part that's because mid-range arrays are getting more scalable and have more of the enterprise-class features.

But to your point on scale-out/scale-up, that's exactly why our P4000 (formerly HP LeftHand) and X9000 (formerly IBRIX) are key to where HP storage is going. The P4000 is a unique clustered storage architecture that allows customers to scale performance and capacity together in an iSCSI-based SAN. The X9000 is similar but for unstructured data. Over the coming months, you'll start to hear a lot more about how these kinds of architectures are central to customers converged infrastructure strategies. Ciao!

Related Posts:

Ricerca

Subscription Options:

Twitter

Post recenti

Archivio

Links

Related Posts: