Australia reveals some information about its Magnus supercomputer
Share on Twitter.
Get the most reliable SMTP service for your business. You wished you got it sooner!
May 12, 2014
Australia's Pawsey Supercomputing initiative has revealed some basic information about an
upgrade to its 'Magnus' supercomputer.
The upgrade in question will see Magnus transform itself into a Cray XC-30 supercomputer with
over 35,000 CPU cores.
Intel's future Xeon processor E5-2600 version 3 will be soon be introduced in service to give Magnus
its upgrade and take it past the petaflop barrier.
The new Intel CPU is due to be announced by the end of Q2 2014, and to have clock speeds of 100 MHz to
200 MHz faster than the current generation of Xeon processors.
Pawsey says that the upgrade means that Magnus “is largely expected to be the most powerful supercomputer
in the southern hemisphere.”
The new Xeon CPUs are certainly giving Magnus a very comprehensive upgrade as it currently possesses 3,328
active cores and runs at 69 teraflops.
That's rather fewer than Pawsey's other super Galaxy, which at 9,440 cores comes in at number 209
on the November 2013 list of the world's top 500 supercomputers.
Getting Magnus to a petaflop will likely mean it beats Galaxy's position on that list.
The supercomputers' ranking list is pretty fluid because new ones are being built all the time.
But with Magnus going large and Galaxy on track to add two more entries alongside its five other
systems in the top 500 before long, things are certainly looking up for the Pawsey Supercomputing initiative.
Overall, Magnus spends most of its time on radio astronomy and geoscience computational issues.
The upgrade to the petascale level will help it do better in both fields, and whatever other
purposes researchers from the iVEC consortium that funds the machine can dream up, the new system
is bound to perform as expected.
In other IT news
IBM has refreshed its SVC (SAN Volume Controller) and Storwize storage array products, providing doubled processing power, boosting compression as well as
increasing Storwize capacity to around 4 PB.
There are two detailed blogs about the new hardware and software, one from inventor Barry Whyte
and the other from Tony Pearson, also an IBM inventor.
Pearson ventures this claim-- “Having sold over 55,000 systems and managing over 1.6 Exabytes of
data, IBM continues to be the Number One leader in storage virtualisation solutions.”
That’s a claim that will be evaluated and probably denied by EMC, HDS and NetApp, and probably HP
as well, but we'll have to see about that.
The Storwize family of products looks like this:
SAN Volume Controller (SVC)
Storwize V7000 Unified
Flex System V7000
The SVC 2145-DH8 follows on from the previous 2145-CG8 product, which came in a 1U X86
server node enclosures, with each needing cache protection as well.
So we now have a 2U device which contains two hot-swap batteries instead of the separate
UPS. It also has a new engine featuring:
Dual socket 8 core Xeon CPU based on the Ivy Bridge core design and additional 32 GB RAM
One or two compression cards with Intel QuickAssist chip to speed compression and replace
previous SW-only compression
The ability to attach up to 48 SSDs via 12 Gigs SAS cables instead of having four smaller
SSDs inside the node as before
Up to six Hardware Interface Cards (HICs) that provide either 8 Gigs FCP, 10 Gigs iSCSI/FCoE,
or SAS plus 3 x 1 GigE ports for iSCSI and admin
Overall, IBM is the first array supplier to bring Intel’s Quick Assist chip to market.
The 48 SSDs are inserted into a separate 2U shelf, based on the new Storwize 12 Gigs SAS expansion
enclosure, which provides up to 38.4 TB of flash storage for an SVC IO group.
The old 2145-CG8 SVC can be upgraded to the new product or it can be added in to an existing SVC
cluster if needed.
Whyte says-- "The DH8 provides around 2x the IOPs and up to 3x the GB/sec of the CG8. Watch out for some new SPC benchmarks to prove this soon."
In other IT news
Yet another company is making a bet that platform-as-a-services clouds are the future and
Amazon Web Services' infrastructure-as-a-service technology is the past.
This time its VMware's subsidiary Pivotal which has launched a "Pivotal Web Services" cloud based
on its own Cloud Foundry platform-as-a-service to take on equivalent services from Google (App
Engine), Amazon (AWS Elastic Beanstalk), Salesforce (Heroku), and Microsoft (Azure).
PWS runs on top of CloudFoundry 6 (CF 6), which is a PaaS service with open source elements. PaaS's
allows developers less control over underlying infrastructure but claim to make application
deployment significantly easier.
CF is written in Go (its predecessor was made in Ruby), and Pivotal says this gives it
greater and faster performance.
Initially though, the PaaS supports apps written in Ruby, Node.js., and Java, but developers
can add their own applications and runtimes by building "Custom Buildpacks".
"Overall, Buildpacks are a convenient way of packaging framework and/or runtime support for
your applications," Pivotal explains.
"For example, by default, Cloud Foundry does not support Python, or a Python framework like Django.
Using a buildpack for Python and Django would allow you to add support for these at the deployment
stage," added Pivotal.
Buildpacks are one of the keys to Pivotal's overall technology strategy, which sees the company
take a different approach from those of its rivals.
Pricing has been designed to be substantially lower than its competitors, with Pivotal's pricing
for its service working out to about $10.80 per month versus Amazon's $35.27.
Though the company's comparison isn't exactly correct as it stands now, as AWS's "Elastic Load
Balancer" has different characteristics to its "HTTP Routing and Balancing", it serves as a good
rough estimate for where the service fits.
PWS also comes with a "Services Marketplace", which means that people can easily integrate
and pay for additional technology such as ClearDB's MySQL Database or SendGrid for email delivery,
or a MongoDB-as-a-Service technology from MongoLab, and so on.
But overall, Pivotal Web Services is a serious bet by VMware that it can succeed in a cloud
world. The main emphasis of the company is on-premises installs wrapped in a pricy package of
It's also worth remembering that we've been here before-- both Google and Microsoft's early cloud
services were both platform-as-a-service systems, and they failed to make the companies significant
sums of money.
Both have since developed infrastructure-as-a-service platforms as part of a Plan B for greater revenue.
Source: Australia's Pawsey Supercomputing Project.
Get the most dependable SMTP server for your company. You will congratulate yourself!
Share on Twitter.
Need to know more about the cloud? Sign up for your free Cloud Hosting White Paper.