'Cheap' Oracle box bashes NetApp benchmark An Oracle mid-range ZFS storage array has beaten a NetApp filer on a SPEC benchmark, despite costing just one-fifth of the NetApp price. The SPECsfs2008 is designed to evaluate the speed and request-handling capabilities of file servers utilising the NFSv3 and CIFS protocols. We're only concerned with NFS here. The Sun ZFS …
This post has been deleted by its author
I was going to say: 'Yes but the ageing FAS6080 is only put in as an extra data point. The main article compares the $179,602 Oracle 7320 with a $1,215,290 FAS3270 which didn't start selling until late last year.' Until I read the comment below. If the 7320 really is just running RAID0 (striped Vdevs), then it's a silly comparison. If you ask them nicely, could they repeat the comparison with raidz2 or, even better, striped raidz2, please?
Where's the "We live in hope" icon? Oh, Paris!
This is ridiculous. You're comparing a deployment spec filer running raid-DP(6) to an Oracle array configured as raid-0:
"The pools are then set up to stripe the data (RAID0) across all 17 drives."
And you fail to even mention it in your article??? Don't you think that's *SOMEWHAT* important information?
The choice of RAID-0 is somewhat bizarre for anyone who actually cares about their data. Still, at least ZFS will tell you *which* files are trashed if you get disk sector errors.
We have not had the pleasure of using NetApp kit to know how well or otherwise they work, though we have heard enough about their usurious licensing arrangements to want to avoid them.
But we have had the experience of Oracle's storage, and let me tell you its a winner in the donkey sucking stakes. Otherwise, it is great. Just so long as you don't have to make *any* change to *any* setting, or expect support to actually *fix* something.
I'm reminded of when Apple made an entry into storage and released a 1U storage array which had IIRC 1TB of storage, they advertised all over the place about how much less expensive it was than HP RAID. They did, however, neglect to mention that HP's devices had the then top of the range SCSI drives and theirs had Ultra ATA.
And .. it gets much worse.
HOW THE FUCK IS ANYONE PROUD OF BELOW ONE MILLION IOPS IN RAID 0 FOR TWO HUNDRED GRAND ???
Seriously, the engineers @ oracle - and the guy who posted this , should reconsider their career in technology - definitely misplaced.
And then to the hardware ...
2 2 10 Gigabit Ethernet Adapter Oracle Sun PCI-E Dual 10GbE Fiber Dual port 10Gb Ethernet adapter
Right here you could have a bottleneck . if the array was faster though ...
4 2 Disk Drive w/Shelf Oracle J4410 SAS 20x300GB 15K,HDD
5 4 Disk Drive w/Shelf Oracle J4410 SAS 24x300GB 15K,HDD
Welcome to retard land, where people still use SAS 15krpm drives ... way to go Oracle - so 2010 though. - and they're expensive obviously, apart from being useless and bad IOPS and fragile ...
6 8 SSD Drive Oracle 512GB Solid State Drive SATA-2 SSD Read Flash Accelerator 512GB
Only @ oracle will you find SATA-2 512GB flash drives being sold in 2012 .. pitiful - and it does matter a lot since cache BW is a major factor.
7 8 SSD Drive Oracle SAS-2 73GB 3.5-inch SSD Write Flash Accelerator SSD Write Flash Accelerator 73GB
Again . wtf ? Oracle .??? 73 GB SAS 3.5 inch SSD ? never heard of actually fast SSD ? you can trash those 8 disks and put a single SLC pcie Accelerator it'll be faster and cheaper. take two for redundancy even ...
So basically, we now know that the hardware stack people @ oracle are just ... tourists.
For exactly the same rackspace, and going for one of the dumbest solutions possible, we could just go ahead and make a reliable ZFS flash-only array that would cost half the price and have double the performance.
"Again . wtf ? Oracle .??? 73 GB SAS 3.5 inch SSD ?"
While your general view of the stupid nature of the test is right, and the impressive nature of SSD for random IOPS, you basically don't understand ZFS and file system journalling.
The choice of a moderate size write-optimised SSD for the ZFS Intent Log makes sense as, for a small amount of hardware cost (compared to a lot of HDD), you *dramatically* increase the ability to do random write operations.
Basically the ZIL deals with the write steps for a few second until the main array (and RAM cache) has committed the writes and the journal can be updated to say so (and in the sensible case of RAID-Z or RAID-Z2 it has done the necessary array stripe read-modify-write operation).
"does not justify the use of 8 crappy SSD's where a raid 1 of fast ssd's would've done the job"
It depends on your requirements. We have over 120TB of storage and, admittedly, a mostly large file / roughly sequential access statistic on several clients. I think we would find that the cost/performance trade off from (in our case) striped RAID-Z2 HDD & SSD on the journal versus the whole array would be interesting.
You care to tell us what the cost for a 120TB SSD array would be?
The NetApp FAS3270 offers overall 4x more storage than the Oracle appliance, btw and has been available since November, 2010 whereas the Oracle gear will not be commercially available until May, 2012.
Oracle threw in loads of memory (the controller alone has 288GB) whereas NetApp had 36GB all in all (I read "a huge cache" can make up for a lower spindle count).
Can anyone provide realistic discount figures for hardware available today from the two vendors?
Memory prices plumetted recently but still, the NetApp's profit margin seems to be heftier. Perhaps they were competing against some EMC apples back in 2010.
Come one, you can do better than this... (fresh) Apples to (fresh) apples please?
@W&W
I agree that Oracle using more memory is not really fair in this bench. But, on the other hand, NetApp costed a million more. If NetApp added more memory, NetApp would be even more expensive. Noone can dispute the big price difference, more RAM or not. If Oracle increases the price one million, Oracle could add much more hardware and other features.
No matter how you see it, the NetApp solution is hefty over priced compared to Oracle. That we can all agree on.
Oracle have always offered powerful and robust solutions. It's the first time I've ever seen them come in as the cheaper option, though!
More seriously though, interesting article, but I would have liked a little more detail about the systems and where these differences come from. Also, where are both products in their life cycle?
That's garbage. How does 37TB of RAID0 storage serving 3 hosts compare with 125TB of RAID-DP storage serving 12 hosts? One of those solutions I would actually put in a production environment, the other is designed to run benchmarks really quickly.
And actually, with that capacity requirement (162TB raw) I would probably use a lesser FAS3210 and flash cache (sorely missing) with 12x600GB shelves and the cost would go down - I would cut it down to 4 shelves total (57TB) to compete with Oracle's pitiful 37TB (and watch the cost drop significantly as well while STILL maintaining hot spares and RAID-DP protection for my data).
I think this article is only written to get storage guys all riled up about a junk comparison more than anything.
It is funny how lot of people complain on Oracle's $200 K server, while the NetApp $1.200 K server receives no complaints. I mean, some people would say that NetApp's pricing is totally off the charts and would complain lot. But instead of complaining on NetApp's gross pricing, people complain on Oracle. On top of that, people implies there is NetApp fud going on, from the author.
What the heck, I can buy several Oracle servers for the price of ONE NetApp, and still no one mentions that. Instead, this article is "anti NetApp"! Let us face it, if NetApp had competetive prices, this article would have been written. Instead, NetApp is an dinosaur with IBM like prices, of course NetApp is frightened and the NetApp people complains on this Unfair comparison. :o)
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
