By mws on Nov 09, 2008
Today I'm pleased to announce the new Sun Storage 7000 Series of unified network storage devices. For the past three years, the Fishworks team have been locked away at an undisclosed location creating this incredible new line of storage devices, and today we're taking the curtain off and announcing them to the world. We'll do it live by webcast from sun.com at 3pm Pacific time.
Three years ago, my long-time partner-in-crime Bryan Cantrill and I created the Fishworks advanced engineering team at Sun with a simple goal: take the core of Sun's innovations in systems and software, and use these innovations to create a breakthrough line of integrated storage products. Our goal was quite simple: create a beautiful product that fully integrates the software and hardware experience (hence our name of F-I-S-H), deliver some killer storage features long missing from the market, and provide a world-class enterprise product at an unprecedented price point. And we've done just that. Bryan recounts the history of Fishworks in his blog, and the rest of the team gears up the rest of the story about our products. For my part, I'll try to provide the big picture of what we did, how we did it, and what comes next.
The New Economics
For more than a decade, storage systems have delivered performance (measured in IOPS for latency or MB/s for bandwidth) through two basic mechanisms: lots of DRAM or NVRAM cache, and lots of fast disks (those 15k RPM FC drives you have in your NAS box). And to really drive IOPS, you go further and take your 15k RPM disks and over-provision the storage, thereby short-stroking the data around the outer tracks. It works, but it's left us with an unfortunate legacy:
- DRAM and 15k RPM drives are the two most expensive things in a large-scale storage system
- DRAM and 15k RPM drives are the two most power-hungry and hot things in the system
- You can't have maximum capacity and maximum performance simultaneously, since those 15k RPM drives are a lot smaller than highest-density 7200 RPM drives
- Drive speeds aren't getting any faster, only drive density is increasing
- Your storage vendor starting monetizing every last knob and screw of the architecture, charging exorbitant software fees and requiring license keys for each protocol and feature you use
- Your storage vendor never implemented a scalable operating system, so storage devices don't scale with CPU efficiently, and are completely off the commodity compute doubling curve
Today we are changing all of that. The Sun Storage 7000 series delivers a completely new economic model for storage, based on the simple premise that using Flash memory, we can build storage performance in an entirely new way, using a Hybrid Storage Pool of transparently-managed DRAM, read-optimized Flash, write-optimized Flash, and low-cost, low-power spindles. Then we put all of that on top of an industry-standard compute architecture that has volume and doubles every eighteen months, with the most scalable storage microcode in the industry, the OpenSolaris kernel.
And then we throw in a dose of reality: no software fees, no license keys. Every bit of storage software we have is included for the price of the box. Then add whatever support plan you need, either from Sun or one of our partners. As one of our beta customers put it, "When I consider price, performance, rack density, and power consumption, your new storage systems give me 16X the storage value per dollar spent." A new economics indeed.
The Killer App
Every new product needs a killer app, and we've delivered one with these new storage products: Analytics. Analytics is a revolutionary new way of observing and understanding what your storage system is doing, in production, using real-time graphics. It lets you take any aspect of the storage system (protocols, disks, network, cpu, memory) and ask an arbitrary question about your workload and get an immediate answer over the web interface while the system is running. Then just point-and-click on something interesting and drill down to ask a new question. It goes like this:
- How many IOPS am I delivering?
- How many for CIFS and how many for NFS?
- What CIFS clients are most active?
- On the most active CIFS client, what files are being accessed?
- For them most active file, can you show me the read-write mix?
You'll have the answer in less time than it took to read the preceding paragraph. It looks like this:
Analytics is the perfect match for our revolutionary Hybrid Storage Pool architecture because we've empowered storage administrators with the ability to understand their workload, and then given them unprecedented insight into how they can grow their architecture to improve performance. Need more networking? Add more read-optimized Flash to your caching hierarchy? Want to understand whether mirroring or RAID-Z DP is best for your workload? Now you'll have the real answer on the only workload that matters: the one running in your datacenter.
The products we're introducing today include:
- The Sun Storage 7110, a 2u box with 2TB of storage,
- The Sun Storage 7210, a 4u box with 46TB of storage, and
- The Sun Storage 7410, a 2u box with up to 288TB of storage that can be clustered (and we'll double that to 576T with a software update in a few months)
All of them have the same software features, except the 7410 adds active-active clustering. And all of the software features come included with the box, including advanced features like replication, compression, thin-provisioning, and all of our data protocols.
The 7410 is the full expression of the Hybrid Storage Pool architecture. It supports:
- Up to 16 cores (32 for a cluster)
- Up to 128G of DRAM (256G for a cluster)
- Up to 600G of Read-optimized Flash (1.2T for a cluster)
- Up to 288G of Write-optimized Flash
- Up to 288T of raw disk capacity
And then you have plenty of PCIe lanes to plug in things like 2x10Gb or 4x1Gb Ethernet cards, or an FC HBA to connect to a tape library for backup.
The FeaturesAll our systems include these core data protocols:
- NFS v3 and v4
- RAID-Z (RAID-5 and RAID-6), Mirrored, and Striped disk configurations
- Unlimited Read-only and Read-write Snapshots, with Snapshot Schedules
- Built-in Data Compression
- Remote Replication of data for Disaster Recovery
- Active-Active Clustering (in the Sun Storage 7410) for High Availability
- Thin Provisioning of iSCSI LUNs
- Virus Scanning and Quarantine
- NDMP Backup and Restore
Finally, to maximize the availability of your data in production, the Sun Storage products include a complete end-to-end architecture for data integrity, including redundancies at every level of the stack. Key features include:
- Predictive Self-Healing and Diagnosis of all System FRUs: CPUs, DRAM, I/O cards, Disks, Fans, Power Supplies
- ZFS End-to-End Data Checksums of all Data and Metadata, protecting data throughout the stack
- RAID-6 (DP) and optional RAID-6 Across JBODs
- Active-Active Clustering for High Availability
- Link Aggregations and IP Multipathing for Network Failure Protection
- I/O Multipathing between the Sun Storage 7410 and JBODs
- Integrated Software Restart of all System Software Services
- Phone-Home of Telemetry for all Software and Hardware Issues
- Lights-out Management of each System for Remote Power Control and Console Access
And a really nice user interface:
To make it easy for you to try out Analytics and get started with these products, check out the home page for the Sun Storage 7000 Series and download our Unified Storage Simulator: it runs in a virtual environment on your laptop but is an entirely functioning network storage device.
Over the next few months, I'll dive inside the implementation details of many of the features we worked on to make these products happen, and talk about how we see the storage world changing. One thing for certain is that today's announcement is only a beginning: we've defined the storage architecture for the next ten years, and we intend to make good use of it. There are three things that will be at the core of everything we do: a true software architecture for Flash, the Hybrid Storage Pool; users empowered with real-time Analytics so you can finally understand what your box is doing and how to make it better; and an open, industry-standard architecture: open on-disk formats, open protocols, and a compute and i/o architecture with volume economics behind it that doubles in speed and capacity every 12-18 months.
Enjoy the launch.