HP’s new P10000 3Par Systems – Exciting details emerging!

In the next few weeks HP will be announcing the P10000 as the latest addition to their storage portfolio of products.  I recently attended a session where a bunch of techies got a sneak peek on some “under the hood” updates which will now be HP/3Par’s flagship storage system.  I seem to be the first person writing about the P10000 on the internet and since no NDA’s were required to be signed I don’t expect any black helicopters following me tomorrow.  Two systems under the umbrella of P10000 will be released, the V400 and the V800.  I am told that these new systems were conceived before HP’s acquisition of 3Par so it seems that little or no HP influence has been applied here.   When it comes to the P10000 HP wants to instill three phrases into your mind, Efficient, Autonomic, and Multi-Tenant.  They are branded as the most powerful virtualized array for virtual and cloud data centers.  The P10000 model designation along with the specifications signifies that this array will sit on top of the HP Storage totem pole.  I put together some bullet points below outlining some new features/updates.

• GEN4 ASIC – Two GEN4 ASIC’s per controller with 1GB chunklet sizes vs. the current 256MB.  The raise in chunklet size was to accommodate modern drives which have a much larger capacity then what was out when the 256MB standard was developed.  With the 256MB standard today’s large drives can have 6000+ chunkles, which creates a lot of work to track due to the sheer number.  Since there are two ASIC’s, one will connect to half of the nodes and the other will connect to the other half to improve workload.  Improvements on zero detect are expected.   Zero-detection finds unused capacity within “fat” data volumes with the ability to provide fat-to-thin volume conversions.
• CONTROLLER CPU – Current 3Par systems have two, dual core processors.  Now, two QUAD core Intel processors will be at the heart of the controller nodes.  I am not sure which Intel series chip will be used but it is definitely a more modern processor because of the additional cores.
• EXPECTED CAPACITY/PERFORMANCE – The V400 will have room for 2 or 4 controllers with estimated conservative performance numbers of 6500MB/S and 180,000 IOPS.  V800 will support 2,4,6,or 8 controllers with performance numbers dialing in at 13,000MB/s and 360,000 IOPS!
• X64 OPERATING SYSTEM – The current x32 OS is a limitation, especially in areas with the amount of cache the system can access.  The new x64 OS will provide much greater scalability and access to more RAM.
• BANDWIDTH – Thanks to the move from PCI-X to PCI-E, and switched PCI-E the arteries of the system have been opened up which results in a huge gains of bandwidth over the previous systems.  The ASIC’s between two controller nodes can talk to each other in speeds up to 112GB/s with virtually no latency.   A 250% increase in bandwidth has been realized.
• ADDRESSABLE CACHE – This was a limiting factor in previous generation so the addressable cache to the CPU has been upped significantly which is probably due to the x64 OS upgrade.   The current T-series has 4GB of control cache with only 3GB being addressable.   The new V400 will have 16GB of control cache, the V800 will sit at 32GB.  Data cache for the V400 will have 32GB, V800 doubles that with 64GB.
• SSD Drive in Node – Since max cache climbs to 96GB per node an SSD drive is now used to dump the node cache quickly enough in the event of the system running on battery backup due to a power failure.
• PEER MOTION – Peer Motion is a cool new feature that is classified as a “data migration and storage federation tool.”  For example, let’s say we want to move a LUN to a new 3Par system and we have our array with a LUN presented to a host.   You will also present that LUN to a new 3PAR destination.  At this point you will see two paths as the destination 3Par presents out to the host using the same WWN’s.  As I/O flows, the original path will drop and I/O will flow only through the destination array.   The LUN migration is Thin’d in the process and all throughout this the host has no idea that this is going on.  3Par is the only supported source and destination arrays right now, but that may change in the future as this is a much more efficient way or migrating systems that what currently out there.  Hopefully compatibility with other source arrays like EVA is in the works.
• T10DIF – Robust CRC-based data protection will be performed in the backend.   Drives will be 520byte formatted.
• RACKS/CABLES – v400 will be rack mountable with standard racks.  Pod Rack compatibility for the v-series may be available later in the year.   Stock 3Par racks are 40U, 36 inches deep with a locking front and back door.  Front to back cooling and 100% of the cabling will be in the rear.
• SERVICE PROCESSOR – Every 3Par system has this, it constantly talks to the array where any problems or alerts will result in a call home scenario.  Every 4 hours snapshot of configuration is taken and every six hours snapshots are taken of performance.  When the system needs to be serviced by a technician they plug into the Service Processor and use the guided maintenance system that states what needs servicing, specific tools required, how to perform the procedure, and using lights to point the tech to the actual part that needs replacement.  This is great for preventing the obvious mistakes that can occur when dealing with these large systems.
• THIN REMOTE COPY – Today if you have thinly provisioned volume and you are replicating to another thinly provisioned volume, writes coming in will detect zeros and ship them across the wire to the target array but will not be written to the back end.  Now, if a write comes in and it is all zeros they will NOT be shipped across the wire and will just tell the other side about the zeros.  This is possibly due to the “WRITESAME” command, but not confirmed. 

The overall 3Par architecture has not changed, which is a great thing.   You still have Thin built into the ASIC, Persistant cache, Fine-grained virtualization, wide striping, mesh-active controller design, and mixed workload support.  The InForm OS (v3.11) and Recovery Manager software have all been revamped.  Expect major integration improvements with VMWare, SQl, Oracle, Exchange, etc.  I am hoping that the software lineup meets, and in some cases exceeds NetApp’s offering.  Firm details will not be known until HP officially releases specifications, so consider this all speculation coming from me.  Anything can change between now and the launch date which should be at the end of the month at which I will post updates as they emerge.  Overall, this is a home run for both the HP/3Par team, this new system coupled with vSphere 5 and HP’s blade offering should be a home run for any cloud implementation!  It would be awesome if we could see this system at VMWorld, come on HP!