Upload
insidehpc
View
222
Download
0
Embed Size (px)
Citation preview
Supercomputing 2016
NCAR Globally Accessible User Environment
DDN Booth
Supercomputing 2016 Pamela Hill, Manager, Data Analysis Services
Supercomputing 2016
National Center For Atmospheric Research - NCAR
• Federally(Funded(Research(and(Development(Center((FFRDC)(sponsored(by(the(Na:onal(Science(Founda:on((NSF)(and(established(in(1959(
• Operated(by(the(University(Corpora:on(for(Atmospheric(Research((UCAR),(a(nonEprofit(consor:um(of(>(100(member(universi:es,(academic(affiliates,(and(interna:onal(affiliates(
“I$have$a$very$strong$feeling$that$science$exists$to$serve$human$welfare.$$It’s$wonderful$to$have$the$opportunity$given$us$by$society$to$do$basic$research,$but$in$return,$we$have$a$very$important$moral$responsibility$to$apply$that$research$to$benefi?ng$humanity.”$
$Walter(Orr(Roberts,(((((((((NCAR(Founding(Director(
Supercomputing 2016
Computational & Information Systems Laboratory – CISL Mission
• To(support,(enhance,(and(extend(the(capabili:es(for(transforma:ve(science(to(the(university(community(and(the(broader(scien:fic(community,(na:onally(and(interna:onally(
• Provide(capacity(and(capability(supercompu:ng(
• Develop(and(support(robust,(accessible,(innova:ve(and(advanced(services(and(tools(
• Create(an(Earth(System(Knowledge(Environment(
Supercomputing 2016
Data Analysis Services Group NCAR / CISL / HSS / DASG
• Data Transfer and Storage Services • Pamela Hill • Joey Mendoza
• High-Performance File Systems
• Data Transfer Protocols • Science Gateway
Support • Innovative I/O
Solutions
• Visualization Services • John Clyne • Scott Pearse • Samuel Li
• VAPOR development and support
• 3D visualization • Visualization User
Support
Supercomputing 2016
GLADE Mission GLobally Accessible Data Environment
• Unified and consistent data environment for NCAR HPC • Supercomputers, Data Analysis and Visualization Clusters • Support for project work spaces • Support for shared data transfer interfaces • Support for Science Gateways and access to Research Data Archive (RDA) &
Earth Systems Grid (ESG) data sets
• Data is available at high bandwidth to any server or supercomputer within the GLADE environment
• Resources outside the environment can manipulate data using common interfaces
• Choice of interfaces supports current projects; platform is flexible to support future projects
Supercomputing 2016
GLADE Environment
GLADE$
HPSS$
Data$Transfer$Gateways$
Science$Gateways$
Project$Spaces$Data$Collec?ons$$HOME$$$$$$WORK$
$SCRATCH$
Remote$Visualiza?on$
Computa?on$ Analysis$&$Visualiza?on$
RDA$ESG$CDP$
geyser$caldera$pronghorn$
yellowstone$cheyenne&
Globus$Online$Globus+$Data$Share$
GridFTP$HSI$/$HTAR$
scp,$sXp,$bbcp$
VirtualGL$
Supercomputing 2016
GLADE Today
• 90 GB/s bandwidth • 16 PB useable capacity • 76 IBM DCS3700 • 6840 3TB drives
• shared data + metadata
• 20 NSD servers • 6 management nodes • File Services
• FDR • 10 GbE
Production Jan 2017
• 220 GB/s bandwidth • 20 PB useable capacity • 8 DDN SFA14KXE • 3360 8TB drives
• data only • 48 800GB SSD
• Metadata • File Services
• EDR • 40 GbE
Expansion Spring 2017 • 19 PB useable
capacity
Supercomputing 2016
DDN SFA14KXE Scalable System Unit
8 x SSU’s • (4) NSD VM’s • (4) Dual Port ConnectX-4 VPI
HBA’s • (4) EDR IB Connections • (4) 40 GbE Connections
• (420) 8 TB NL-SAS • (6) 800 GB SSD
(42) 8 TB
(42) 8 TB
(42) 8 TB
(42) 8 TB
(42) 8 TB, (1) SSD
(42) 8 TB, (1) SSD
(42) 8 TB, (1) SSD
(42) 8 TB, (1) SSD
(42) 8 TB, (1) SSD
(42) 8 TB, (1) SSD
(2) NSD VM’s
(2) NSD VM’s
EDR IB
40 GbE
Supercomputing 2016
GLADE I/O Network
• Network architecture providing global access to data storage from multiple HPC resources
• Flexibility provided by support of multiple connectivity options and multiple compute network topologies • 10GbE, 40GbE, FDR, EDR • Full Fat Tree, Quasi Fat Tree, Hypercube
• Scalability allows for addition of new HPC or storage resources • Agnostic with respect to vendor and file system • Can support multiple solutions simultaneously
Supercomputing 2016
HPSS Archive160PB, 12.5 GB/s
60PB holdings
HPSS
GLADE16 PB, 90 GB/s, GPFS20 PB, 220GB/s, GPFS
(4) DTN
Data Services20GB/s
FRGP Network
Science Gateways
10 Gb Ethernet 40 Gb Ethernet 100 Gb Ethernet Infiniband
100 GbE
Data Networks
Infiniband FDR
Geyser, Caldera Pronghorn 46 nodes
HPC / DAV
HPSS (Boulder)Disaster Recovery
HPSS
Cheyenne 4,032 nodes
Infiniband EDR
NWSC Core
Ethernet
Switch 10/40/10
0
10 GbE 10 GbE
40 GbE
40 GbE
Yellowstone 4,536 nodes
Infiniband FDR
Supercomputing 2016
GLADE I/O Network Connections
• All NSD servers are connected to 40GbE network • IBM NSD servers are connected to the FDR IB network
• Yellowstone is a full fat tree network • Geyser/Caldera/Pronghorn quasi fat tree, up/dwn routing • NSD servers are up/dwn routing
• DDN SFA’s are connected to the EDR IB network • Cheyenne is an enhanced hypercube • NSD VM’s are nodes in the hypercube
• Each NSD VM is connected to two points in the hypercube • Data transfer gateways, RDA, ESG and CDP science gateways are
connected to 40GbE and 10GbE networks • NSD servers will route traffic over the 40GbE network to serve data to both
IB networks
Supercomputing 2016
GLADE Manager Nodes
• Secondary GPFS cluster manager, quorum node
glademgt1
• Primary GPFS cluster manager, quorum node
glademgt2
• token manager, quorum node, file system manager
glademgt3
• token manger, quorum node, file system manager
glademgt4
• Primary GPFS configuration manager • token manager, quorum node, file system manager, multi-cluster
contact node
glademgt5
• Secondary GPFS configuration manager • token manager, multi-cluster contact node
glademgt6
Supercomputing 2016
GLADE File System Structure
GLADE Spectrum Scale Cluster
scratch
SSD metadata
SFA14KXE 15 PB
8M block, 4K inode
p
DCS3700 10 PB
4M block 512 byte inode
p2
SSD metadata
SFA14KXE 5 PB
8M block, 4K inode
DCS3700 5 PB
home apps
SSD metadata
SSD applications
SFA7700 100 TB
1M block, 4K inode
• ILM rules to migrate between pools • Data collections pinned to DCS3700
Supercomputing 2016
Spectrum Scale Multi-Cluster
• Clusters can serve file systems or be diskless • Each cluster is managed separately
• can have different administrative domains • Each storage cluster controls who has access to which file systems • Any cluster mounting a file system must have TCP/IP access from all nodes to the
storage cluster • Diskless clusters only need TCP/IP access to the storage cluster hosting the file
system they wish to mount • No need for access to other diskless clusters
• Each cluster is configured with a GPFS subnet which allows NSD servers in the storage cluster to act as routers • Allows blocking of a subnet when necessary
• User names and group need to be sync’d across the multi-cluster domain
Supercomputing 2016
Spectrum Scale Multi-Cluster
Geyser, Caldera Pronghorn 46 nodes
Cheyenne 4,032 nodes
Yellowstone 4,536 nodes n1 n2 n4536
n1 n2 n46
nsd21
nsd22
nsd54
nsd1
nsd2
nsd20
ESG / CDP Science
Gateways
Data Transfer Services
RDA Science Gateway
web n1 n8
n1 n2 n6
web n1 n8
EDR Network
FDR Network
40GbE Network
. . .
. . .
. . .
. . .
. . .
. . .
Storage Cluster YS
Cluster
SAGE Cluster
RDA Cluster
DAV Cluster
CH Cluster
DTN Cluster
n1 n2 n10 CH PBS Cluster
n1 n2 n4032
Cheyenne PBS, Login Nodes
. . .
. . .. . .
GLADE
Supercomputing 2016
HPC Futures Lab - I/O Innovation Lab
• Provide infrastructure necessary for I/O hardware and software testing • Focused on future I/O architectures and how they might relate to NCAR HPC • Proof of Concept for future HPC procurements • Development environment for upcoming I/O environment changes • Domains
• High-Bandwidth Networks (40/100 GbE, IB, OPA) • Parallel File Systems, early release & pre-production testing (Spectrum Scale, Lustre, ??) • Storage Hardware (block storage, object storage, SSD, NVMe, ??) • Hadoop Cluster (testing of local temporary storage) • Cloud Storage Clusters (targeted for Data Service Gateways) • Compute Cluster (to drive I/O tests)
Supercomputing 2016
HPC Futures Lab - I/O Innovation Lab
• Boulder Colorado Lab • Network Infrastructure to support 40/100 GbE, EDR IB,
OPA • Ability to add additional network technology as
necessary • Management Infrastructure
• Cluster management tools • Monitoring tools • Persistent storage for home directories and
application code • Storage and server hardware
• Cheyenne Wyoming Lab • Leverages current test infrastructure
• Production network support for 40/100 GbE, IB • Integration with current pre-production
compute clusters • Focused on ‘Burst Buffer’ and NVMe environments
• Easily coupled to test compute clusters
Supercomputing 2016
I/O Innovation Lab – Boulder Lab
Parallel File System Research
Compute / Hadoop Cluster
Cloud
Cloud
Cloud
Cloud
40GbE
iolab mgt1
iolab mgt2
40GbE
40GbE
OPA
EDR GPFS Mgt
FDR
GPFS NSD
GPFS NSD
Lustre OSS
Lustre OSS
Lustre MS
Lustre Mgt
SSD
SFA 7700 Storage
40GbE
EDR
OPA
Hadoop
Hadoop
Hadoop
Hadoop
Hadoop
Hadoop
Hadoop
Hadoop
Cloud Data Services Research
Lab Infrastructure
Supercomputing 2016
I/O Innovation Lab – Cheyenne Lab
IBM GPFS
IBM DCS3700
DDN GPFS
DDN SFA14KXE
IBM Jellystone
SGI Laramie
IBM yogi/booboo
DDN IME
DDN IME
SGI UV300
SGI NVMe
40GbE
FDR
EDR
GPFS I/O Routing from FDR to EDR networks
Connected to EDR at FDR rate
Leverages picnic infrastructure
Leverages current test systems for I/O
testing
Supercomputing 2016
QUESTIONS? [email protected]