Slashdot Mirror
Sun Unveils RAID-Less Storage Appliance
pisadinho writes "eWEEK's Chris Preimesberger explains how Sun Microsystems has completely discarded RAID volume management in its new Amber Road storage boxes, released today. Because it uses the Zettabyte File System, the Amber Road has eliminated the use of RAID arrays, RAID controllers and volume management software — meaning that it's very fast and easy to use."
Correct me if I'm wrong, but doesn't charging enterprise prices for simplified hardware that relies on commodity software solutions, kind of defeat the point?
Unless I'm misunderstanding this hardware, the entire idea is to move data safety away from hardware redundancy toward software-driven duplication. In that way, the data is safe from failure in the same way that GoogleFS protects against individual machine failures. The only difference is that Google probably doesn't pay $11,000 for 2TB of storage. :-/
One of these days, I really will understand why Sun regularly shoots themselves in the foot. Until then, I suppose I must trust them to somehow find a customer who's willing to pay exorbitant prices for an otherwise good idea. (i.e. I'd really love to see Sun bring Google-style reliability from unreliability to the market.)
BTW, here's the link to Sun's marketing on this:
http://www.sun.com/storage/disk_systems/unified_storage/index.jsp
It's actually pretty cool tech. Sun could own the market if they just understood how the market views pricing and features.
Javascript + Nintendo DSi = DSiCade
"All of the new unified storage systems include comprehensive data services at no extra cost, Fowler said. These include snapshots/cloning, restores, mirroring, RAID-5, RAID-6, replication, active-active clustering, compression, thin provisioning, CIFS (Common Internet File System), NFS (Network File System), iSCSI, HTTP/FTP and WebDAV (Web-based Distributed Authoring and Versioning)."
Note that this system includes "RAID".
Considering that they've purchased MySQL, StorageTec and Cluster File Systems (of Lustre fame), developed ZFS, implemented CIFS in OpenSolaris from scratch (not Samba based), participated in NFSv4 and constructed the thumper, these machines hardly come as a surprise.
For the last two years, almost all their moves are targeted towards one goal: Enter the storage market from a non-conventional angle. They want to do it unconventionally, because they know that storage more than anything else is becoming The commodity and today's toys won't cut it. Plus, at this point, all the mainstream storage vendors have difficulty tapping the low end. They may be able to sell their expensive products to clients with deep pockets, but for small businesses it's a different story. No to mention that they are unwilling to reinvent themselves. OTOH with all these inventions Sun may be trying to do what it did with workstations when it started in the 80s, start low and increase. Remains to be seen whether they can pull it.
This system will intelligently move the data around to put frequently accessed bits on the SSDs. This is a lot more than a 2u server with a few TB drives in a raid 10.
You don't make the poor richer by making the rich poorer. - Winston Churchill
ZFS doesn't stand for zettabyte anything. "The name originally stood for "Zettabyte File System", but is now an orphan acronym." from wikipedia, sourced from http://blogs.sun.com/bonwick/entry/you_say_zeta_i_say .
and of course "RAID Array" is lovelily redundant phrasing.
People, please stop trying to compare a couple of drives from Newegg tossed in a chassis as a similar product for thousands less, simply because you have the same storage capacity.
That's not even apples and oranges, it's more like apples and redwoods.
Last I checked Netapp was still charging $10,000 per TB! Do you really think there is no reason for this?
Experience teaches only the teachable. -AH
Will that $600 box be using 14 146 GB 10k RPM SAS disks?
These boxes aren't about providing stupid storage, their about providing massive I/O throughput. The larger boxes scale to 44TB and 576TB respectively. This also automatically moves frequently accessed data to flash drives (and RAM) for even faster I/O.
These are absolutely monstrous compared to anything you could build for $600. There seems to be quite a bit of custom hardware to power this setup.
You don't make the poor richer by making the rich poorer. - Winston Churchill
The third one I believe--the rest I'm skeptical about...
"Not an actor, but he plays one on TV."
With that said, linux REALLY needs ZFS , and not in userspace.
Due to deliberate licensing issues we won't have native ZFS in Linux any time soon. However, BtrFS should be merging into the mainline kernel soon enough (~2.6.29), and it includes most of ZFS's features plus a few of its own: storage pools, checksumming, mutable snapshots, built-in extent-level striping and mirroring, etc. It even supports in-place, reversible conversion from ext3 via a copy-on-write snapshot.
"The state is that great fiction by which everyone tries to live at the expense of everyone else." - Bastiat
"Sun replace RAID with RAID"
No, they replaced it with "RAD"; they took the "I" right out of it.
+0 Meh
Yes yes yes, you can do that with just $1000 and a afternoon at Fry's or browsing Newegg, right?
Everyone's missing the point here, and a lot of what is being said could be applied (just as wrongly) to NetApp... after all, those are just x86 boxes running a BSD kernel.
The special sauce here is not so much the underlying OpenSolaris OS (which does provid the IO and services such as CIFS, NFS, iSCSI, data replication, and so on) but the Fishworks software put on top of it. Built-in failover clustering, the integrated web GUI and CLI... if you weren't paying attention to the console during boot, you might not even have a clue that's it's OpenSolaris underneath... which is one of the marks of a good appliance OS... easy to manage and the idiosyncrasies of the underlying OS is sufficiently abstracted away.
You don't need to be a Solaris admin to use this, just like you don't need to know about BSD to run a NetApp. The difference here is that this takes pretty high-end x86 hardware and does better than NetApp, for cheap. Ever see a support contract for any of the NetApp filers? I guarantee you'll spin a 180 on your heals and pretend you never saw the number.
While it certainly doesn't make up the ~$50k difference on its own, the 7210 _does_ come with 64G of RAM (vs 16G) and a pair of 18G SSDs. They're not completely identical.
Fortune 500 companies typically standardize hardware, so people who say they can buy this from here, that from there, one more thing from eBay are rediculous.
Also, to those who say small businesses can't afford this, its really an option. Some places like open source hodgepodges of hardware and some do not because their small business generates enough money that investing in enterprise class hardware with gold 4 hour response from a solid company with a history of UNIX experience and integration with Solaris.
Also, said Fortune 500 companies get massive discounts, as what you're seeing is retail price.
It supports active directory, and user mapping between AD and LDAP. The CIFS stack is in-kernel.
But SUN is FAR from being the inventor of charging people $50k for something they could just as well get for free...
Name ANY big IT vendor, they all do it. My father can tell some amazing stories on that subject. Not a new phenomenon either.
Now, if you are the GOVERNMENT, they'll give you the special bonus public sector price, $150k!!!
"Malo periculosam, libertatem quam quietam servitutem." -- Jefferson
to me. Coming from high performance transaction processing land where an operation means 'the data is ON the platter' you can't do that more often than the platter rotates to the point where the head is over the sector where the write operation starts. Basic math, 15k RPM spindle = roughly 300 times/sec. Multiply by however many spindles you got, that's what you're max throughput is.
This is one reason why IN THEORY at least an SSD would be so great, that latency is much less. So basically I'm thinking they just aren't talking about what you're talking about, and maybe that makes sense, if you're running a trading operation say, you just DO NOT CARE what is buffered someplace, if it isn't physically on the drive, it doesn't exist.
"Malo periculosam, libertatem quam quietam servitutem." -- Jefferson
Guh. Sorry. I'm tired, and re-reading my comment the english is well-formed but the concepts are jumbled nonsense. Let me try again, by your leave...
Yes, it's unavoidable to rebuild when you lose a disk, and there will be a performance hit unless you go for full on 100% redundancy, and not many companies can afford to do that with a lot of data.
ZFS offers a number of benefits, though, in the event of drive failure-triggered rebuild, in that it basically knows where the data is and only bothers with that. A hardware controller has no idea what's data and what is blank space and so just redoes everything. In theory, assuming the MB/s of rebuild is the same, a ZFS rebuild of a half-full array should take half the time of a traditional controller.
It is also much more intelligent about *what* it rebuilds, starting at the top and then descending down the FS tree, marking it as known good along the way. This means that if a second drive fails halfway through the resync, instead of a catastrophic failure you still have the data up to the point of failure.
I can't remember where I read that; maybe here: http://blogs.sun.com/bonwick/entry/smokin_mirrors
But I didn't even want to talk about drive-failure rebuilding, what I actually wanted to say that ZFS is, in theory, less likely to get itself into an inconsistent state in the case of power fluctuations, controller RAM failures, drive failures w/ pending writes, that kind of thing. That's the kind of rebuild I meant - after some kind of catastrophic failure. I should probably have said "integrity checking" though.
By design, ZFS never holds critical data in memory only and so at least in theory should always be consistent on-disk. Basically it shouldn't need to fsck. That is a giant advantage to me, if it turns out to be as good in reality as it sounds on paper. Of course, that also has a lot to do with the capabilities of the FS proper, but removing the evil, evil HW controllers from the picture can only be a plus.
I don't know why, but RAID controllers are the most unreliable pieces of hardware I have ever known, besides the drives themselves (but at least they are consistent and expected to fail). Get a few of them together and something WILL go wrong, more often than not in a horrible and unexpected way. When some RAM goes bad in a HW RAID controller you are in for a whole lot of subtle, silent-error-prone fun. Anything that gets the HW controllers out of the picture is a win for me.
And don't even mention the batteries in HW raid controllers. They are the wrong solution to the power failure problem, especially since it's always after a failure that a disk will decide it's had enough of spinning and would just like to sit still for a while, thank you very much. Drive failure with pending writes! Exactly the words every administrator wants to hear. Almost as good as power failure with pending writes. Combine the two (highly likely!) for maximum LULZ. Ok, this is turning into a rant, I better stop.
Anyway, thanks for the corrections. My original comment (and probably this one) came across as a confused mess upon re-reading .. sorry .. will sleep now : )
Let my new 7-digit UID be a lesson to all - write down your passwords.
No, they replaced it with "RAD"; they took the "I" right out of it.
It's all fun and games until someone loses an "I".
The goal of this product is to compete with Netapp. If you've ever experienced Netapp licensing/pricing, this Sun solution is a bargain. People seem to be forgetting that this is a storage appliance.
This is meant to be 100x faster than the storage you're talking about:
First: This uses Hybrid Storage Pool:
The Hybrid Storage Pool combines DRAM, SSDs, and HDDs in the same system, dramatically reducing bottlenecks and providing breakthrough speed.
Second: The system's hybrid architecture gives you the speed and performance you need to shatter the I/O bottlenecks with no administrator intervention. In fact, Hybrid Storage Pools with SSDs can improve I/O performance by 100x compared to mechanical disk drives.
It's funny how this viewpoint is always the one promoted on slashdot. One could argue that the Linux GPL is the problem. FreeBSD and Mac OS X had no problem integrating ZFS into their code precisely because the ZFS license (CDDL) allowed it.