IBM today announced a complete redesign of its System x x86
server family, featuring up to 12.8 Tbytes of NAND flash directly on
the memory bus of the server and a modular design that allows users to
upgrade the server by simply replacing plug-in “Compute Books”. This
sixth generation of System x has six core reference architectures
including one for SAP HANA and is designed to facilitate the
virtualization of ERP and other large, core enterprise-level
applications for delivery through private and hybrid clouds. The
scalable design can also reduce acquisition costs up to 28 percent in
comparison to competitive Xeon x86 systems, IBM says.
The announcement includes a new System x3850 M4 BD storage
server, a two-socket rack server supporting up to 14 flash and/or disk
drives delivering up to 56 Tbytes of high-density storage. This server,
which combines compute with storage, is specifically designed for large
build-out architectures such as Hadoop big data installations, says Stuart McRae, IBM System x high-end business line manager.
It also includes the new IBM FlashSystem 840, providing double the
bandwidth and performance – 1.1M IOPS – of its predecessor, the
FlashSystem 820. It supports up to 48 Tbytes of usable capacity in a 2U
unit with IBM Microlatency
technology that cuts data access times to microseconds. Designed to
support big data, it provides actionable insights from real-time data
analytics faster than its predecessors. It also features a new
management GUI and datacenter-optimized features such as hot-swap
components and concurrent code load, enabling fast installation and
easier management.
Virtualize your ERP for the cloud
“The System x6 is the first server family that’s been
effectively designed from the ground up to incorporate flash storage,”
McRae said. “Until now, flash storage has been kind of an add-on – you
add on a PCI card to the server. This is integrating flash storage on
the memory bus, the highest speed bus in the system, and making that
available as a block storage device, that looks like any other block
storage device to the application.”
By putting the flash on the memory bus, it becomes the
fastest flash storage on the market. And these new systems can support a
lot of it.
“It looks exactly like a DDR3 DIMM,” he said. “These
systems are going to have 92 DIMM sockets in a four-way, so it can
support up to 6 terabytes of system memory on a four-way, or 12
terabytes in an eight-way. That’s three times as much memory as is
available in a standard eight-way server today.”
This has major implications both for big data analytics and
for virtualization of very large enterprise applications such as ERP.
“If you wanted to cache a five terabyte database in the server for
analytics applications, you can configure that as a cache.” So for
instance, in the SAP HANA
appliance, a large amount of that space is used for RAM, allowing users
to have a very large data set in HANA while still providing large
amounts of flash for staging data. And by spreading a HANA or similar
installation across several servers, it can support very large databases
on the memory bus while providing resilience and redundancy in case of a
hardware or power failure in any one server.
Before the x6 generation, users were memory-constrained in
what they could virtualize. “If you had a four-way server only
supporting one or two Tbytes of memory, it’s hard to virtualize a
terabyte application,” McRae said. “Now they can do that on these new
platforms.” This opens the way for virtualization of ERP systems, even
Oracle Red Stacks, that today run on bare metal, allowing customers to
realize the advantages of server virtualization and deliver services
based on their ERP and other core systems to users via their private
clouds.
“I want to move my large databases to a cloud model. I want
to move my SAP HANA to a cloud model. I want to move my big ERP
applications. I don’t want to have to re-architect it to a new
architecture, I want to move it now, and this provides the
infrastructure to do that,” MacRae said.
“Booking” your memory, flash, CPUs
The
other part of the x6 revolution is the new parallel modular design that
IBM calls “Compute Books”. Each server is made up of these
plug-and-play modules, each with its own processor and memory. These
plug into a backplane that provides power and IO.
This means that upgrading a server or replacing a failing
unit is simply a matter of unplugging one or more modules and plugging
in replacements. Then a simple restart implements the new hardware
without requiring a forklift replacement and all the management that
goes along with that. MacRae says IBM estimates that the core server
will support at least the next three generations of processor and
memory/flash technology.
“Once you’ve architected the server and put your big
applications on it, two years from now, when you say ‘Scotty, I need
more power,’ you just pull the Compute Books out and plug the latest,
greatest ones in. It’s all transparent to the back-end IO.” That
provides a great deal of investment protection across generations.
And while it does require a reboot of the upgraded server,
“because it’s a virtualized environment, and now we’ve virtualized these
large applications, you have no application downtime.”
Six reference architectures
As part of the announcement, IBM also announced six
pre-architected versions that come with software installed: an SQL data
Warehouse, a Hyper-V appliance running on Windows Server, an SAP HANA
version, an SAP Business Suite version, a VMware vCloud, and finally a
version running DB2 with BLU acceleration on Linux. The servers come
with either SUSI or Red Hat Linux or Microsoft Server. While IBM does
not have a reference architecture for it, Oracle has System x on its
compatibility list, so users can also run an Oracle Red Stack on the new
System x. And because of the higher end processors and the large
amounts of memory and flash storage that the new generation supports,
they can decrease the number of licenses they need, saving significant
cost, particularly with Oracle. And System x also runs IBM Watson for
users who want that in-house rather than using it from IBM’s cloud.
By: Bert Latamore
Aucun commentaire:
Enregistrer un commentaire