Design your own Global Name Space
I met a customer
today where they look for a true global filesystem.
The customer has
today a data center in Sweden and need to open another data center in north
america.
They customer have
today a calculation cluster based on open source. To be able to share data
between their calucation nodes are they using NFS share today.
Based on regular NFS
and the few functions they are using today did they see problems with either
the NFS cache wasn't syncronized with the source or they client nodes access
time to the source file was to slow.
Behind the source
NFS Server was a regular midrange storage system from a large global disk
vender.
And they even try
with the vendors version of "flash system" that is SSD and not true
flash, but still it is to slow.
The customer has
today approx 1 PB of data and yearly are they growing with 2,5PB of data and
each calculation batch is approx 40-80TB data per calculation.
The data are
normally only used once or maybe twice during the data life time, and the life
time per 2,5PB of data is 5-10 years.
What we at Cristie
suggested was a IBM Spectrum Scale File System and a Spectrum Scale Server.
To be able to get a
quick and cost effective solution did we choose does products to be able to
have a full flexible and open filesystem that the customer can grow with any
kind of hardware and still use easy building block solution such Spectrum Scale
Server.
All archive data are
we saving to a IBM TS4500 Tape library and using Spectrum Archive solution.
Instead of buying a
true flash system or a Spectrum Scale Server with only SSD, did the customer
ask for a less cost solution.
Because of the
Spectrum Scale Server GL6 generates much better IOPS and GB/s then there
existing solution will we only need a much smaller SSD cache then the customer
first thought.
Because we are using
Spectrum Scale Servers, do we need a server based gateway that we choose to run
Linux, to be able to manage Spectrum Archive.
We did then fill up
all server with maximum numbers of cheap SSDs from Lenovo, and to get better
performace and fail-over cluster functionallity did we choose two nodes to a
single GPFS SSD pool.
Now does the
customer have a full Spectrum Scale solution using Disk, SSD and tape, and all
working together as one single namespace filesystem. The customer don't need to
use NFS Cache or regular slow NFS connetions anymore.
The customer get a
true global software define filesystem, that is flexible and scalible with any
hardware.
Now can do they have
a solution they can expand easly in north america, but also expand the global
namespace to other countries such Sweden, by using a software such Aspera to be
able transfer data between the sites can they then save a lot of money buy not
investing in a large network infrastructure.
Together with
Cristie World Wide delivery functionallity did the customer make a success
implementation, by sending the hardware to another country but still been
invoicing in Sweden.
Comments
Post a Comment