16 posts • joined 23 May 2008
> and, who knows, object storage with its non-RAID data recovery features like erasure coding
RAID is an implementation of erasure coding.
> may make a desktop entry, either as the storage layer underlying the top-level file access or as a direct offering for apps with built-in object storage access.
What compelling reason would there be to putting an object filesystem layer in here for desktop use? It won't speed up rebuilds.
There are a few ways to speed up RAID rebuilds:
* Smaller raid sets (though at a cost of capacity)
* Efficiently rebuilding by removing as many bottlenecks as possible (more drives, controllers, SAS lanes etc) - this is what most people think when they're talking about erasure coding but equally applies to single arrays.
* Only rebuild data, not empty space. This is also what most people thinking about erasure coding schemes are referring to, but while that's impressive on an empty array it's less so when the array is 100% and all 'enterprise' arrays should do that anyway, if they own/understand the filesystem and hence know what is empty space.
To capacity. The bit error rates on these drives should give you pause if you aren't using at least dual parity. 64TB is 5.12e14 bits... close to the BER 1e15/TB the drives are rated for. Smaller RAID sets help - e.g. 5+0 is better than 5 for this case, but 6 is better than 5+0 for the same number of drives if resilience is needed.
Only reason to have RAID-5 on this would be for scratch space just so that a single bit error didn't cause a long running multi-hour job to fail. RAID-0 just isn't any use at all with these drives.
> NetApp FDAS
A new model?
I looked at the numbers. 200MB/sec and low IOPs ' from published materials' but nowhere does he reference the model, age, configuration, benchmark aim, price, or even just a few URLs
> EMC VNX & NetApp FAS not recommended for primary storage workloads involving high random IOPs
Seriously? I'm guessing that someone needs to tell both ours, and EMCs, massive customer bases running practically every workload possible that they've obviously made a mistake. Or could it be that we're comparing Apples to Oranges? What are we comparing here?
Next I looked for externally audited benchmarks from Permabit on SPC-1: http://www.storageperformance.org/results/benchmark_results_spc1/#spc1
So I'll disclose now and I must put up my blog post. On a special benchmark I ran on my lab at NetApp I got 420,000,000,000,000 IOPs and 6PB/sec! Beat that!
Re: Netapp dedupe is a scam
Stacker was a data compression technology. NetApp dedupe is taking 4KB blocks that are duplicates and eliminating all but one of them. NetApp also has compression as well.
NetApp gives compression & deduplication as part of the base license (i.e. no additional cost).
Please explain the scam further since I work at NetApp and am not aware of it.
I think I understand your point. If you want just one specific capability or feature then the rest do not matter. I agree.
Some customers buy devices for 4-5 years and have an expectation that parts of their business or technical requirements may change in the future - e.g. the application gets upgraded and now recommends different ways of deploying (Hyper-V/SQL over SMB3 as a recent example), or it turns out that some new way to deploy something unlocks capacity that can be used elsewhere, etc.
If change is a remote possibility then having a flexible storage architecture that can adapt and still continue to work the same way will likely save substantial money/effort/pain.
Re: Unified architecture...
As well as the protocols is the ability to have the same feature set across them regardless of size of platform. Want to take a snapshot? Mirror the data? Clone something? Enable space efficiency for a set of LUNs or Files? All work exactly the same. Want to use the smallest machine with a few drives? Or the largest with multiple PB? That is the power of a true Unified architecture.
LUNs in ONTAP are first class citizens, no NAS emulated LUNs or anything like that. You talk natively to the array using LUN semantics and it reads/writes data to disk.
WAFL was originally developed for NFS, so when we added CIFS as a first class protocol you would say that it has an emulated CIFS layer on top of NFS? We don't translate CIFS connections into NFS, and we don't translate LUN calls into NFS/CIFS. We do take all of those protocols and use WAFL to efficiently write them to disk (via RAID, another abstraction layer).
Now, why would it matter anyway? I'm not sure I understand your point. Other vendors may indeed use emulation. If it works, has performance (e.g. good SPC-1 results), and generally acts like a LUN, I wouldn't care if they emulated & virtualised (Virtual SAN type LUNs?), or had craftsman hand-carving each bit lovingly into the disk. Why would a customer care? Price, performance, features, stability of company, partnerships, vision. Those are likely much more important.
Better than A-SIS?
Disclaimer: I work at NetApp.
Ignoring the rest of the speculation in the article since it was an entertaining read, I noticed this:
> NetApp would gain the DXi deduplication software which is a stronger technology than its own A-SIS deduplication.
Is it really stronger? I'd be interested to hear about any proof that the DXi technology has much value in primary production workloads which is where a lot of A-SIS is being used. I would suggest that being able to deduplicate production workloads is stronger than deduplicating static backup data.
NAS vs SAN or "cheap array" vs "expensive array"
> the key difference to focus on is whether or not you need the top performance and reliability of a SAN and are prepared to pay the premium. If not, you need a NAS.
NAS vs. SAN misses the point. The quality of the array and requirements of the host OS and software is much more important than the colour of the network cable, unless you are looking at the ultra-low end where a $100 NAS box will be much worse than a $10000 SAN box.
At the mid/high end, a high quality array will do NAS and SAN equally well, and will also allow apps to run over both classes of network equally well (and in most cases, at the same time).
BTW - someone needs to tell Oracle or VMware that NAS is not performant or reliable.
I work at NetApp
I'm not the AC. I do work at NetApp.
> Um... OnTap 8 or 7 Mode
Is it broken? If so, how? (BTW, as a certified NetApp individual, you'll know that ONTAP 8 comes in 2 versions, cluster-mode and 7-mode. It isn't "ONTAP 8 or 7-mode", its "ONTAP 8 7-mode or ONTAP 8 cluster-mode").
> OSSV not working outside of GMT +8 / -8?
Seriously? Please point me at the burt / NOW page that says this... since we have an entire APAC business that would struggle to use it...
> RAID-DP is not RAID 6.
According to whom? SNIA, the industry association created by the vendors to represent all of us non-politically, says it is: http://www.snia.org/education/dictionary/r/ (RAID-DP is diagonal)
> averare of 15% loss in real read performace with a NetApp vs. EMC
What? Evidenced how, under what workloads? With what configurations?
> Or howabout the fact that PAM/FlashCache is READ ONLY?
Is this a problem? Why is this bad? [hint, for a NetApp array with WAFL and NVRAM, it's the right way. For other vendors, not so much]
> EMC FASTCache is R/W!
Great. What does this mean? Why is this good? An answer that includes evidence such as published, audited benchmarks would be even better...
> How about "flash is pointless.... oh we've got flash now, it's good" or "Object Storage sux... we bought bytecast, now it's good" or "Not Compresss.... no....wait, we've got it now..
Unified Storage? NAS?
We've been selling Flash for a long time - PAMII / Flash Cache is flash... (hence the name...)
Object Storage is one of those "could be a gigantic market in the future so invest now" but enterprise apps such as Exchange, Oracle etc don't support it, so for most customers it is just an interesting discussion point.
As for compression time-to-market (and sales strategy)- we aren't perfect, and it's just icing on the storage efficiency cake.
> SnapShot's???? Why, oh why do my snapshots occur in my expensive production disks?
For performance and efficiency. That is why Snapshots do not hurt performance, unlike some of our competitors arrays. It's also why dedupe works exceedingly well since we can use the blocks that are also in the snapshots.
> Oh thats write, snapshots can only occur in the same volume... not like EMC where I can snap shot to much, much less expensive disk.
Expensive is a relative term since dedicating entire drives to your snap space, having to have spares of those drive types, hoping that you got the ratio of "production" vs. "snapshot" disks correct, and the performance drop for using them, is probably not worth the effort. Then restore (the reason for them in the first place); if you have to restore an entire multi-tb database from a snapshot it is much quicker if you don't have to move data around to do it...
> Now for a bit of a reality check:
Can we have something more modern than 2007; if this is really a reality check then I assume there will be tons of data to support all of your assumptions.
> So here are the FUD Rules (again):
FUD Rules? I thought the first rule of FUD Rules is not to talk about FUD Rules?
Don't compare NFS vs CIFS results
In the article:
...and returned a throughput score of 18,784, 2.36 times less than the CIFS throughput score. As a crude comparison, let's suppose then that the EMC Celerra Gateway NS-8 set-up above would score proportionately the same in the NFS benchmark.
From the SPEC website referenced in the article:
SPECsfs2008 results may only be compared to other SPECsfs2008 results for the same protocol. SPECsfs2008_cifs and SPECsfs2008_nfs.v3 are not comparable because they are generated using completely different workloads.
So, no, it's not a crude comparison. You can't just do maths and assume that the platforms perform identically with different protocols. Because none do. NFS (2/3) are much simpler protocols than CIFS and it can show on the throughputs on devices that support both (natively). Where non-native CIFS emulation (e.g. Samba) is involved I'd expect the difference to be orders of magnitude different.
Maybe it's a virtualised tablet that you run on a normal computer? You could use the mouse to "touch" the screen. That would be pure win!
£50 per hive per year? 18 hives? £900 lost, plus the bees of course. For a £6000 investment that's a pretty poor ROI given that you've got to maintain the hives, prepare and ship the honey.
Paris, 'cos she's in it for the love too.
If that's the best of the rest then I'll stick with my Touch...
> After all this is a public service and not there to line someone's pockets!
Hmm... another thought
£450m in cost.
60m or so people in the UK.
That works out at about £7.50 per person.
Wow. Nice to know how my tax money is being spent.
Surely there is a better way of doing these things though. We aren't having a general election this year so the people who could have been doing that could be doing this instead. Or tracking down the people who have been affected by the 10p tax removal (select * from population where income<18000)?
And don't do these things every 10 years if it's a big deal. Do approximately 1/10th of the house holds every year based on NI numbers so that people only get bothered every so often.
Why does it all just seem so hard?
Shame we can't do this stuff in Britain
It's amazing we have to continually outsource everything. The government shouldn't need to outsource it's own data collection and processing. And by outsource I mean outsource to any commercial company, foreign or domestic.
- Analysis iPhone 6: The final straw for Android makers eaten alive by the data parasite?
- First Crack Man buys iPHONE 6 and DROPS IT to SMASH on PURPOSE
- TOR users become FBI's No.1 hacking target after legal power grab
- Vid Reg bloke zips through an iPHONE 6 queue from ZERO to 60 SECONDS
- Analysis Why Oracle CEO Larry Ellison had to go ... Except he hasn't