Sun FlashFire's record-breaking storage performance will make network admins giddy
By Tim Stevens 
posted Oct 14th 2009 8:33AM
posted Oct 14th 2009 8:33AM
posted Oct 14th 2009 8:33AM
Now that we've thrown 'em off the trail, use the form below to get in touch with the people at Engadget. Please fill in all of the required fields because they're required.
I bet it can't compare to Microsoft/Danger's crashing abilities!...
BURNED!
Now if only there was a way to convince the NCO to buy this.................................
and it will cost 11 billion $ :)
thats not so bad considering what the company gains i.e massive usage sites like facebook or system mainframes like the US government could boost productivity a great deal with this monster in tow
Where is the real story/news-release? I would like to see the block size, etc. to get 1.2M I/O.
We had an NDA presentation a few months back about these things. The IOPS numbers (obviously) come from small block IO (~4K). The main drawback, however, is the inability to get those numbers out of a single large volume. These things have several SAS controllers that can't talk to each other. As a result, you have to carve out a minimum of 4(?) LUNs. Those 4 LUNs have to be equally distributed over the available SAS links to the host AND be utilitized at 100% each to get the peak perf. Not that you could really find a single host that'll push those numbers, but it's still an annoying draw-back.
All that aside, the arch is pretty neat. Super capacitors for power instead of batteries -- super fast charge, just enough capacity to flush IO in a power failure. We would have loved to get one of these to test with Lustre -- aside from the minimum of 4 LUNs, these are ideal for Lustre metadata storage. 4k IO, and super-uniform seek times and multipath capable -- who cares about capacity?
Oh, another thing. This took a while to get to market because Sun fabbed their own flash modules for this. They're basically SO-DIMMs with a lil sata(?) controller onboard. They wanted to be as cooling and space efficient as possible. You can get it in 20, 40 and 80 DIMM configurations (and naturally performance will depend on how many you get).
http://www.sun.com/storage/disk_systems/sss/f5100/
Google is probably faster
You're an idiot.
I doubt it... they only buy commodity hardware in bulk, and cluster the hell out of it.
I'd like to point out Dataram's XcelaSAN.... That SCREAMS and while it might not be as fast as this new Sun thing it keeps all your data on your existing SAN, so all that replication, cloning and snap shots that you use your SAN for, stay working... I'm working for a company that is one of the beta sites for Dataram, and the thing ROCKS! Screw RAM SAN.
you're talking about two distinct product categories here. The xcela thing is purely a caching appliance and relies on cachable data. This sun product is a storage array.
I understand. My point is that if you want high performance. you can get the XcelaSAN and NOT move your data off of your existing SAN. If you buy this Sun box, you will need to either replace your current SAN, or add this to your existing environment.
the XcelaSAN lets you Accelerate any LUN from your storage array you want, yet keep everything on your existing array.
If you have an EMC Clariion and buy this sun box for your oracle data, you have to move all your Oracle data over to the sun box. If you do this, all your replication and other things the EMC was doing, can no longer be done to the Oracle data.
If you instead get an XcelaSAN and connect it to your CX, you keep everything on your back end SAN, keep your replication etc... AND get a MAJOR performance boost.
they are different solutions to the same problem.
Yes, but you are assuming all data benefits from caching. That might be true for db environments where a small set of queries are constantly made but in a great number of environments, this is not the case and you're just inserting latency into your network with one of these cache boxes. Beyond that, caching does little if anything on the write side. These Sun boxes bring the speed directly to the array, no matter the data type or uniformity of IO.
They are not different solutions for the same problem, they address two very distinct problems.
I agree with you to a certain extent.... The XcelaSAN has improved our write performance significantly. It allows us to write significantly more to the XcelaSAN then we could to our SAN and the XS sends the ack immediately. Then the XS keeps data in cache for reads, & offloads it to disk ASAP.
I would say there are overlapping areas where the XS would benefit instead of just getting a new array like this SUN box. And others where it wont help.. But the XS advantage is its much cheaper then getting a new array yet still yields a major performance boost in a large majority of cases.
It doesn't help us much for backups, but for most of our DBs and much of our other SAN data it helps immensely.
So what if you got about 50 of these in differant countries and massive data pipelines connecting them? Mobile Pirate Bay?!?
How would it be mobile...?
well the data would be moved from one place to another if there is a problem with the law at some place. this would be totally transparent for the user.
I can burnz rubber on information super highwayz?
Not to nitpick, but wouldn't this scare the ether out of the network admin, and make the storage admin giddy?
I guess it depends on the company. In smaller ones they're probably the same person.
That's what I'm thinking. It'll make the sysadmins excited, network admins terrified, and DBAs will say "oh, actually, 2TB isn't enough, could you add another 2TB by tomorrow? kthxbye."
This would have huge benefits to a smaller group of servers segmented off on their own that would pound this device for all it's worth. No need at all for the entire network's bandwidth be reduced for this.
Considering even 10Gb isn't common to the desktop, these would have to be in a SAN, since you'd want at least 103Gb + overhead. This thing would just about max out PCIe x16 if it was directly connected to the bus, say nothing of any intermediate tech!
Bolts to Bits implementation for Oracle here we come!
Something tells me this might end up in Oracle's benchmark reference platform for RAC.
HooYah!
It already has - see Sun Oracle Database Machine (http://www.oracle.com/database/database-machine.html).
actually, I'm pretty sure network admins don't really give a hoot about some silly storage technology. You might be thinking of system admins, or more likely DBAs.
And really, this thing isn't that impressive, sure its fast, but its small and likely very expensive. Just becaue they threw expensive disk technology at the problem doesn't make it innovative, or even a good idea.
In a way this is akin to suping up that old '73 dodge charger. Sure, you can drop a bigger enginer in there, bore it out, wider tires, etc. But if your goal is anything other than showing it off at tradeshows and online, well you'd be better off with a corola or something.
It's actually pretty innovative in my opinion. Other companies are just sticking flash in their systems and using them as disk. Sun is using the flash NOT as a disk, but as another level of cache in the system, sitting between main memory and disk. So users don't have to decide which datasets need to be fast and move them to flash. The flash just speeds everything up. No changes to how the data is managed.
It's quite innovative. The HPC sector (specifically truly high-perf file systems) have been begging for something like this. File Systems like Lustre and GPFS really love to have separate storage for metadata. The capacity requirements are very low but the IOPS requirements are through the roof. If you can pull off in 1U what literally takes a rack or two worth of spinning disk to achieve, you're going to have high-end customers banging on the door in seconds.
Those benchmarkes are all good and such, but how fast can it delete all the records? Thats the new trend these days.
I bet it needs some serious cooling.
Like a nuclear powered cooling system, somehow.
You're aware that nuclear power plants generate electricity by boiling water to produce steam, right?
yes, that was the joke
Hmm, will the hackers be able to pop in, read massive amounts of data quickly, and be able to get out much faster?
idataplex
Now they need to come up with a way to connect it. That's almost as fast as PCIe x16, and any interconnect will have to be low-overhead to pass as much of the PCIe 16GBps as possible, since this would use 12.8GBps all by itself.
Unless they create a DMA connection for it. But then, DDR3 maxes at that same 12.8GBps. DDR4 will be needed to really do anything with this.
It has 4 independent SAS controllers -- those numbers are aggregates for sure.
4x SAS @ 6Gbps = 24Gbps = 3GBps.
This needs 12.8GBps - note the big "B" - and more if the connection involves any overhead.
The Apple iPod is so much better.
Look at the iPod. Sleek, clean, awesome, and app full.
And what is this?
Some crap?
12GB/s? PFFT. My iPod 2nd Gen can do 15GB/s. You don't see me getting in to the news?
And let's compare the price tag:
iPod: $250.
Junk Storage: Millions of dollars.
Face it guys. iPod wins this challenge hands down.
It wins everything and anything you throw at it. I guess if Sun EVER decides to release an ACTUALLY good competitor to the iPod Touch, then we may have something to base this on.
But come on.
iPod wins again.
And will in about 50 years time as well. I could probably emulate this crap on my iPod ;o
loljk guys.
Dx
This guy speaks the truth.
o_O !?
Really?
Are you for real...or what?!
....okay...pass it this way...it has to be some good a$s shit!