* Posts by random_graph

15 posts • joined 19 Nov 2009

OpenIO wants to turn your spinning rust into object storage nodes

random_graph

"Any Application"?

There's a chicken and egg problem here. By definition, object storage has taken advantage of local file-systems on X86 hosts to store chunks across naive disk drives. This provides the advantage of portability and the economies of "off-the-shelf" hardware. So capacity balancing, healing, namespace management, tiering, etc are all handled at the solution/cluster level. Are designs like Isilon, Cleversafe, Caringo, etc really going to deprecate any of this to the drive component? The only real advantage I see at the system level is that the "host" mostly goes away as a fault-domain. But the huge disadvantage is the economy of scale penalty for these low-volume drives. Naturally Seagate wants margin dollars to shift from hosts to drives, but is it enough to be compelling at the system level when the system still needs to support host-style nodes??

Oh and no freaking way would this HTTP-based solution support "any application". Talk about throwing credibility out the window.

1
1

You're only young but you're going to die: Farewell, all-flash startups

random_graph

The Better Mousetrap rarely wins

I tend to agree with Mellor.

The external storage market is optimizing along the guardrails:

One path: Arrays for transactional workloads, with the purchase led by $/IOPS and solution integration. This is NVME, AFA, and memory-based solutions.

The other path: Storage for persisting unstructured data cheaply and efficiently. This would be scale-out NAS and RESTful object.

And the middle - all the run-of-the-mill enterprise apps running in ESX - will be slowly captured by VSAN/Hyperconverged (including HDFS), especially as VmWare and MSFT duke it out for the full storage stack, blowing away the remaining mid-range array vendors in the process.

Yes Tintri and Pure make a great mousetrap, but in a maturing market, that can never be relied on as the ticket to survival. Customers not going to VSAN/HC will take the low-risk path of EMC, HP, IBM because their products are "good enough". Pure and Tintri, in an effort to show positive cash flow, will eventually have to cut back on cost-of-sales, and then getting the n+1 customer gets harder and harder.

2
0

Scale-out storage: Proprietary? Commodity? Or both?

random_graph

Hardware independence and other myths

A good survey.

The instant these systems target mixed workloads, they sacrifice a lot of capability. In particular, the need to support latency-sensitive, high-transaction-consistency workloads through a block interface, and POSIX-compliant distributed lock management in a file-interface, designers cross a road that they can't come back from. Ceph is a good example; by targeting the generalized cloud workloads, they've had to invest in optimizations for both transaction latency as well as throughput.

The cost is that these systems become bound by rigid assumptions of underlying hardware and network topology. Maybe you'd be asked to choose from the HP Xeon 2u 12 HDD server or the Dell 2u Xeon 12 HDD server, but true hardware heterogeneity is mostly a myth. Dependencies that that inhibit flexibility include drive failure detection relying on specific BIOS versions, specific local file systems, assumptions about how namespace is balanced, intolerance for performance variation, consistency in SMART APIs, etc. Even the promised 'multi-generational' platform support usually results in bulk migrations akin to a forklift upgrade.

Lots of pundits are predicting continued polarization of the storage landscape. Where transaction performance matters, enterprises continue adoption of AFA and eventually NVMe. And at the other end, unstructured data stores go to Object Stores that are optimized for RESTful semantics, WAN/hybrid topology flexibility, yet can provide high throughput (don't confuse with latency) where needed for workloads like batch analytics.

And to answer your primary question, scale-out has proven to deliver improved TCO at scale, due to the simplified administrative experience and ^improved^ platform lifecycle management. But is scale-out a commodity? of course...IMO there ever was a time that scale-out commanded a price premium. Even in the extreme performance HPTC world where Mr. King plays, storage is mostly purchased by the pound.

0
0

Could Rozo squeeze into the scale-out NAS-object scalability gap?

random_graph

Not a better mousetrap, maybe a better spring

There are so many things that make this story incomplete. Although fast erasure coding is a neat trick, there's no reason to believe that Isilon or Qumulo's (or even Rozo's) performance is going to be dramatically bottle-necked or accelerated by the EC calculation.

When I see so many generalizations and superlative claims (performance, scale, etc) and denigrating all things that have come before, I rapidly become skeptical. If they're indeed courting the market where people care about individual file-systems >20PB and massive performance, that means technical HPTC; The customers are going to be seismic processing, national surveillance initiatives, and national labs (NCAR, LLBL, LHC, etc). In this tiny market space they'll have to deal with feature-rich incumbents, parallel client implementations (GPFS, Panasas, Lustre, pNFS), exotic interconnects, and extreme price sensitivity. There's a massive opportunity cost of going down that path, as illustrated by Panasas's difficulty bridging into the enterprise, and Avere's walking away from the NFS caching market.

Everyone in the NAS space likes rendering and DNA sequencing ("commercial HPC") as targeted use-cases because feature requirements are relatively lean, the workload is deterministic, and buyers are anxious to protect themselves from over-dependency on Isilon. But it's a crowded space that already includes new entrants such as Peaxy. With Isilon/EMC slogging through M&A mud, Qumulo is probably best positioned to take share over the next decade. Hopefully Rozo can take their better spring, build a better mousetrap around it, and demonstrate at the solution level they really have some differentiation.

1
0

Corrective lenses needed for Gartner's flashy array vision

random_graph

Artificial compartmentalization misleads buyers

It would be great if we could wave this off as an artificially compartmentalized product ranking by some irrelevant blogger. But Gartner carries legitimate influence with buyers, and this MQ sends the wrong message.

Customers buy arrays to meet a price-performance-capability envelope, and because 95% of all workloads' data (including OLTP) follow the 10/90 hot/cold rule, custoemrs should *always* put some thought towards tiering whether inside the array (Dell, etc) or outside the array (VIPR etc) for maximum ROI.

Purpose-built AFA's could deliver intrinsically higher performance due to less baggage in the data path, but that's not an argument for the artificial classification, it's an argument for a performance-based ranking!

1
0

SAN-free, NAS-free? Scottish PHDs lift kilt on how they'll pull storage out of the aether

random_graph

Oh man, not another one of these

This thing has tried and failed and the past. Noobaa.com is another one, fortunately at least with the good sense to go after the consumer space rather than commercial.

One thing about P2P is that one must compensate for flaky and non-persistent end-nodes. 4x may sound like a lot of redundancy, but what happens when all the employees go home for the evening? Whoops! Even with things like erasure coding, P2P overprovisioning always needs to be prohibitively high. Try convincing that user he needs to contribute 100GB to get 25GB of p2p capacity (yeah but it's FREE!)

And to use this as a VSAN built out of WAN-connected laptops? Yeah good luck convincing that the CEO whose Exchange server just died that saving $25K by skipping that array purchase was the way to go.

1
0

Flash data storage: Knocking TAPE off the archiving top spot

random_graph

Multiple use-cases, multiple architectures

There are 3 drivers for long-term retention of digital content

1) I want to reuse, repurpose, and re-license (eg movie studio)

2) I want to analyze (eg Amazon and Facebook)

3) I want to preserve because that's what I do (US-LOC) or because I promised (Shutterfly)

Each brings different sensitivities in terms of performance SLA, cost, and data integrity. For use-cases 1 & 3, tape-based latencies are entirely acceptable as long as sequential performance is good enough for bulk data operations. Analytics will almost always need more consistent SLAs.

In all cases however, the placement, maintenance, and performance expectation for *metadata* is much more aggressive than the SLA and placement rules for asset itself. For these sorts of storage solutions, the query has always dominated as the first step in data IO, although to-date most often performed against a host database. In the future, storage solutions optimized for these use-cases should recognize that semantic through distributed indexing mechanisms implemented in flash. And although the LTFS file-system may be is useful at a component level, it is inadequate for the task at an aggregate solution level.

0
0

Death by 1,000 cuts: Mainstream storage array suppliers are bleeding

random_graph

It's those dang VmWare guys

Diversity in networked storage solutions was always a function of workload diversity. But with virtualization reaching 100%, the Hypervisor is the only client that midmarket storage vendors need to design for these days. Of course there are still differing performance profiles for different guest workloads, but the point is that the host environment (networking, provisioning, protocols...) has gotten much simpler. So these 3 things represent the bulk of all cannibalization of the legacy array market:

1) Practical adoption costs for using public cloud are much reasonable once you're virtualized

2) Startups can optimize strictly for the virtualized use-case (thus creating differentiation)

3) The hypervisors can handily disintermediate the value chain (EVO-Rail!) and put everyone else out of business (or force them into OEM servitude)

So what's left for the storage vendors? Probably just the vertical workloads (Web2, M&E, HCLS, GIS, analytics, PACS, etc etc).

0
0

It's ALIVE: Unstructured data upstart whips out data-AWARE array

random_graph

Interesting, Just not sure it's solving a real problem

Some interesting new things here, just not sure there's a large market for the unique capabilities. And without the unique capabilities being leveraged, it looks functionally like Isilon, just 13 years late. It seems like text search (not attribute), native audit trails, filter-based tagging, and data demographics are the unique things here. Every SysAdmin wants better data demographics and solid audit, but how many SysAdmins care about Ediscovery or content search? That would be the LOB owner or General Council most likely, people accustomed to host-side tools already. Another surprise is the commitment to proprietary hardware; you'd think the SDS buzz would drive them in the opposite direction. Guess it's that EqualLogic DNA in the founders.

Definitely not the first system to do cool metadata tricks or full-text index (see HCP, Caringo, Honeycomb...). The coolest thing they're doing that I see is by reverse-engineering the VMDK file format they extract embedded file metadata and allow individual file restores. The tagging is also cool. All bets are off though if they're not using the file interface. Unless they install a hypervisor plug-in, they can't get any intelligence out of the iSCSI interface.

0
0

Dell's new Compellent will make you break down in tiers... of flash

random_graph

Re: dedupe block or file based?

The technology is variable block, sliding window plus compression, implemented in the file-system. So it happens at the file layer, but definitely not SIS. And yes developed by the Ocarina team in Santa Clara.

0
0
random_graph

Posr proc vs in-band

We disagree with your assessment Anonymous. As you know Dell already ships products with in-band dedupe and compression, so this is not a technical barrier. In-band is an appropriate implementation for secondary storage. In primary storage, demand histogram invariably follows a 10%/90% (or even 1/99) rule. So like tiering and caching strategies, FluidFS implements a more elegant data reduction design that aligns with the information lifecycle, applying more aggressive savings to the 99% yet maximizing performance for the 1%. And of course your policy-settings are adjustable. [Dell storage dev]

1
2

Looking closely at HP's object storage: Questions Answered

random_graph

What is actually implemented here?

It looks to me like Storeall is merely the marriage of Autonomy and the iBrix file system. But there's no reason to believe the integration goes any deeper than physical. They are merely running index, search, and the RestFul API on the iBrix segment servers. Ibrix is no more object or metadata aware than a Netapp filer.

In regards to what constitutes "Object", There's no reason Object and NAS need to be distinctly defined. The most popular object storage in the world is Isilon, which uses Reed Solomon and distributed placement algorithms with self-healing. Ok, the fact that it's accessed through a heirarchical FS puts it in the NAS camp. But even the file system interface can be merely an expression of object metadata. For example, Sun's Honeycomb could provide an NFS export whose heirarchy was purely a virtualization of a metadata schema...N file systems could be exported with N arrangements of the schema.

I fully expect to see future NAS (and SAN) leveraging more and more object capabilities, and the availability of RESTful APIs representing "the new converged" regardless of underlying architecture.

0
0

Does anyone really want to embed dedupe code?

random_graph

Ocarina & Permabit

Great coverage on the respective issues and possible outcomes Chris.

Permabit and Ocarina are both thinking along the right lines; dedupe is a killer technology for primary storage, and it will increasingly be an embedded feature.

Permabit's comments about Ocarina seem a bit out of place though, "...but our technology is mature and being delivered..."

- Permabit was formed 9 years ago to make a better disk storage system. Dedupe was an feature added onto their object storage system <1 year ago...Now in the latest redo, they're throwing out the product to try to create success on the feature. Ocarina on the other hand started life focused 100% on data reduction for primary storage.

- Permabit's dedupe is <1 year old, and it's never shipped integrated with anything other than Permabit's storage box. Ocarina has been shipping for 2 years, and *every* delivery was integrated with someone else's storage system (Ocarina doesn't store the data).

- Ocarina delivers dedupe *and* compression (a big portfolio of algorithms at that), with proven results on over 1500 file types, including pre-compressed data and specialized data sets in different industries. Furthermore, the end-to-end strategy that Ocarina is talking about is really a next-generation architecture. Permabit's feature-now-product has a long ways to go in technical sophistication to catch up to Ocarina.

0
2

Ocarina compresses Flash and MPEG2s

random_graph

Lossy & Lossless - both available

[I work for Ocarina]

Thanks for everyone's comments.

Just to clarify what we're doing; for web-distributed file types (GIF, JPG, FLV{h.264}) we optionally use lossy techniques. Some lossy opportunities we use are reduction of non-visual info (eg huffman table optimization), spatial optimization (aligning DCT quantization with the HVS), better macroblock management, motion compensation, and more intelligent VBR. Our intent is to never introduce visual artifacts and we even have some portrait studios (making big before/after prints) who have validated the algorithms. With this 'visually lossless' approach, we keep the files on disk in their native format so customers can capture benefits not just in storage savings, but also bandwidth reduction and page-load-time improvements.

For production workflows where loss generally isn't desired we apply a fully bit4bit lossless workflow and use all the proprietary compression we can for maximum reduction. For ingest formats like DV we can get 50% or more. For MPEG2 we're seeing around 20-30% at Beta customers...enough to be meaningful for say a broadcaster's archival system.

And we definitely don't rig the tests ;-) the results are based on customer data-sets only. We work across a thousand file types (so far) and no one here has time to craft a bunch of application-specific data-sets from scratch. Results will vary from customer to customer, and someone who is a real codec expert can almost certainly approximate our results on a specific file type. But we find in practice people don't do that, and that still doesn't provide a scalable dedupe & compression platform that also works on all the other 100 file types in a given customer's workflow, and integrates well with their existing storage system.

I wrote the white paper on Native Format Optimization that talks about the visually lossless approach. I think you have to fill out a form to get it, but you can check it out at www.ocarinanetworks.com

0
0

How can the storage industry prevent cloud bursts?

random_graph

Cloud Storage means everything to everyone

Given the wide variety of definitions for the term, the intersecting justification for the technology can easily be made to be zero. Here are the propagating definitions:

1) Cloud = any web service where a user's data is retained (incl Facebook, Goog docs, etc)

2) Could = storage in a utility-based pricing model ($/GB/month)

3) Cloud = a storage technology with 'cloud' attributes; scalable, self healing, low cost, extra failure resilient, either implemented within an enterprise or over the Internet.

The last definition is the key one...Whether the storage platform is internally hosted or externally hosted, the customer requires technical due diligence and transparency. If they don't get it, they'll choose to walk away! Thus storage vendors (including SSPs) will learn that transparency and SLAs are a key requirement to maintain market share.

Everything regarding a) where it is hosted, and b) how you are charged for it are simply details of implementation, and a function of the providers' business models.

0
0

Forums