Oracle bigs up 7000 ZFS storage line
Mystery top-end product
Oracle has supercharged its 7000 unified storage line, effectively replacing two of the products and expanding the line's performance and capacity dramatically upwards. Users will get more scalability and more performance from the new line.
The new products have been rebranded too, and are known as the ZFS Storage Appliance line with four models: 7120; 7320; 7420; and 7720. The previous line had three models: 7110 scaling to 4.2TB; 7310 scaling to 192TB with a 2-node cluster option; and 7410 scaling to 576TB and again having a 2-node cluster configuration. Both the 7310 and 7410 have up to 600GB of read flash buffering and could have optional write flash buffering too. In the CPU department the 7310 could have up to two 6-core 2.2GHz Xeon processors while the 7410 could have up to four. That's the background against which the 7x20 line has been launched.
John Fowler, Oracle's Sun-inherited EVP for server and storage systems, said: “The next-generation Sun ZFS Storage Appliance product line delivers more than 2x the scale of the previous generation and increases integration with Oracle’s business-critical applications.”
The entry-level 7120 has from 12 to 120TB of capacity using 3.5-inch, 7200rpm, SAS drives. It is a single node system, with a 2U controller and 4U disk shelves, and has a 96GB write cache but no read cache. Oracle calls it an entry-level system, implying it replaces the 7110 with its now meagre 4.2TB max capacity, but that system uses 10,000rpm, 2.5-inch SAS drives and Oracle could continue offering it as a small but high-performance entry-level box, although it has no read or write flash caching.
The 7320 scales to 192TB, exactly the same as the 7310, but the 7320's read flash capabilities are radically better, with up to 4TB of read flash vs 600GB for the 7310. In the CPU department the 7310 has had a slight brain transplant to form the 7320 with the original two by 6-core 2.2GHz Xeons per controller changing to an 8-core processor complex running at 2.4GHz with up to 72GB of RAM vs the 7120's 64GB. The 7320 has the same 1U node enclosure and 4U disk shelf as the 7310 and, obviously enough, replaces it.
The 7420 is a much, much bigger beast than the 7410. It scales to 1.15PB, a little more than twice the 7410's capacity, and can have up to 32 cores running at 1.8 or 2.0GHz in its controller. The 7410 could have 24 cores spread across four 2.2GHz processors. Main memory has doubled to 512GB per node. The 7420's read flash is up to 4TB, a massive increase on the 7410's 600GB, and the write flash has gone up to 96 solid state drives (SSDs). This baby can process many, many more I/Os than the 7410 and, we reckon, replaces it. It comes as a 3U controller node with 4U disk shelves, of which there can be 24v in total, meaning a full configuration would need more than two racks.
Oracle positions the 7410 as the best price/performance box in the line, but there is another 7x20 system, intended for high capacity and data growth rates and numbered the 7720, implying it should be bigger than the 7410. It isn't in a capacity sense, scaling up from 240TB to only 720TB, using 2TB, 7200rpm, SAS drives, as the other 7x20 products do. It can have the same main memory of 512GB as the 7410, the same 4TB of read cache but fewer write flash SSDs - 24 per rack compared to the 7410's 96 maximum. In the CPU department it is slightly ahead, having 32 cores running at 2.26GHz, so where is the real meat in this sizzle?
First, clustering is standard rather than an option. The 7320 and 7420 have active-active 2-node clustering. Secondly, the 7720 is delivered as a rack, with the maximum number of CPUs and back-end I/O capabilities. OK, that's nice, but where's the 7720's real edge over the 7410? We're told the 7720 Storage Appliance provides 1.6 times better storage density than other unified storage solutions. Okay, but that's more sizzle, not steak.
Oracle's blurb (pdf) says the 7420 is "is uniquely designed for virtualised environments requiring multiple data services, and heterogeneous file sharing," and has "extreme scalability". The 7720 "features ease of deployment and industry leading storage density that is designed to fulfill the needs for the largest bulk storage and backup requirements". Only it doesn't - the 7420 holds half as much data again.
Oracle says the 7720 "also serves in mid-scale mixed workload environments". You could get pretty much the same capabilities by having a 2-node, 720TB 7410. It might appear that the 7720 should, in fact, be called the 7420 and the 7420 re-christened the 7720. Did Oracle get the product numbers muddled up?
If the 7720 simply has as better packing density than the 7420 then why is that? Why doesn't the 7420 use the same packing density?
Oracle's 7200 storage documentation says the 7720 "will initially scale to 720TB". Ah, what's going on here? If the 7720 gets higher-capacity drives in the future then they could also be used on the other 7x20 boxes. That can't be the only point of having the 7720 - can it? Is this product, with its slightly greater processing capacity than the 7420, going to be used beyond the 2-node cluster environment, say in 4-node clusters? Alternatively Oracle is going to give it an expansion rack full of disk drives to take its capacity up past the 7420's. For the moment the 7720 is a mystery product that appears to have few, if any, significant features beyond the 7420.
Next page: Back to Earth
Bandwith, there plenty
Each disk shelf has 8 x 6 Gbit connection to controller, which is a lot compared to any other disk system on the market. So bandwith / disk drive is extremely high. while this isquite important for some applications like DWH, it's rarely important to virtualised environments, as IPS and low latency are the first priorities.
7720 uses slighlty beefed up 7420 controllers integrated into an intelligent rack which provides two sides disk slots (back and front), SAS-2 backend, coling and power supply system.
First rack in 7720 system can get 320 data drives + 24 write cache SSD's totalling to 344 drives.
This is a JBOD'less system. Each pre installed storage node takes 32 drives + LCC cards. 2 of the drives can be 0-2 write cache devices.
Hope this helpes
drives and cooling
It's interesting you should mention rear-mounted drives. I initially was thinking that, until I looked at the shelf chassis (it's full-depth), which made me re-consider the drive math. The data sheet does mention that the 7720 will use special cooling. Perhaps the rear Cooling Door? I also thought they might use 2.5" drives to get the increased drive count. It does definitely seem like the final design has yet to be attained. Maybe that explains the slightly later release than the other models.
I think you will find the 7720 cabinet has the disk shelves mounted both front-ward and rear-ward facing....hence the claims of better density than the 7420 on a rack basis. They must be still working out how to scale a 7720 across multiple racks. I do not know how the cooling would work in hot-aisle cold-aisle datacentre.
These devices were obviously completed prior to Brian Cantrill and Adam Leventhal leaving the original 7000 product development team...so it will be interesting to see whether this slows the innovation in these devices (and bug fixing).