Title: The Mass Storage Challenge for the Experiments at LHC: the Solution Developed at INFNCNAF
1The Mass Storage Challenge for the Experiments at
LHC the Solution Developed at INFN-CNAF
IEEE NSS-MIC, Orlando 28 Oct 2009
A. Cavalli1, S. Dal Pra1, L. dellAgnello1, A.
Forti1, D. Gregori1, L. Li Gioi1, B. Martelli1,
A. Prosperini1, P. P. Ricci1, Elisabetta
Ronchieri1, V. Sapunenko1, V. Vagnoni2, R.
Zappi1 1 INFN-CNAF, Bologna, Italy 2 INFN,
Sezione di Bologna, Bologna, Italy
2Overview
- INFN-CNAF
- Mass Storage Challenge
- SRM layer
- Solution I
- StoRM with GPFS
- Results
- Solution II
- StoRM with GPFS and TSM
- Results
- Conclusions
3Overview
- INFN-CNAF
- Mass Storage Challenge
- SRM layer
- Solution I
- StoRM with GPFS
- Results
- Solution II
- StoRM with GPFS and TSM
- Results
- Conclusions
4INFN CNAF computing center
- CNAF is the main INFN computing resource center
- for Large Hadron Collider (LHC) experiments at
CERN Alice, Atlas, CMS, LHCB - .... and many others
- e.g., BABAR (SLAC), CDF (FNAL), VIRGO (Italy),
ARGO (Tibet), AMS (Satellite), PAMELA
(Satellite), MAGIC (Canary Islands) - offering the following resources
- CY 2009 23K HS06, 2.8 PB of disk space and 2.65
PB of tape space on line (2 libraries) - CY 2010 68K HS06, 8.5 PB of disk, 6.6 PB of
tape space on line
5High Energy Physics computing
- Computing at Large Hadron Collider is based on
multi-tiers models - Tier0 (CERN), Tier1s and Tier2s form WLCG/EGEE
Grid - INFN-CNAF is a Tier1
- Uniform access policies to computing resources
(CPU, storage) have been defined in the framework
of Grid (WLCG/EGEE)
6Overview
- INFN -CNAF
- Mass Storage Challenge
- SRM layer
- Solution I
- StoRM with GPFS
- Results
- Solution II
- StoRM with GPFS and TSM
- Results
- Conclusions
7Mass Storage Challenge
- Several PetaBytes of data (online and near-line)
need to be accessed at any time from thousands of
concurrent processes - Aggregated data throughput required, both on
Local Area Network (LAN) and Wide Area Network
(WAN), is of the order of several GB/s. - Long term transparent archiving of data is needed.
8Overview
- INFN-CNAF
- Mass Storage Challenge
- SRM layer
- Solution I
- StoRM with GPFS
- Results
- Solution II
- StoRM with GPFS and TSM
- Results
- Conclusions
9Storage Resource Manager
- In WLCG/EGEE an abstraction layer, called Storage
Resource Manager (SRM), is defined in order to
let client applications interact with the backend
storage system via a common interface (either
disk or tape). - SRM transparently supports several access
protocols for LAN (e.g. POSIX, RFIO, DCAP) and
for WAN (GridFTP). - SRM allows remote space management and
preparation of files for read/write access.
10A typical simplified workflow exploiting SRM
services
- A remote user wants to analyze a dataset stored
on tape - Check whether files exist, e.g. perform remote
file listing by using the SRM service - Order the SRM to move data files from tape to
disk if needed - Remotely poll, asking the SRM what is the status
of the tape recalls - When all files have been recalled to disk, submit
the analysis job - The analysis job contacts again the SRM in order
to get the protocol needed to access the files
and as well as their physical location - Finally, once files have been processed, the
analysis job contacts the SRM in order to release
the files - They can e.g. be deleted from disk through a
garbage collector mechanism if disk space is
reclaimed for other activities
11Overview
- INFN CNAF
- Mass Storage Challenge
- SRM layer
- Solution I
- StoRM with GPFS
- Results
- Solution II
- StoRM with GPFS and TSM
- Results
- Conclusions
12StoRM
- StoRM is an implementation of the SRM solution
designed to leverage the advantages of cluster
file systems (like GPFS) and standard POSIX file
systems in a Grid environment developed at
INFN-CNAF.
13StoRM with GPFS
- GPFS by IBM has been chosen at CNAF as the
solution for disk-based storage demonstrating
outstanding I/O performances and stability - Large GPFS configuration with StoRM as SRM layer
set up at CNAF - 2 PB of net disk space (to be doubled in 2010)
partitioned in several GPFS clusters - 150 disk-servers (NSD GridFTP) connected to
the Storage Area Network (SAN)
14Some results
15Overview
- INFN-CNAF
- Mass Storage Challenge
- SRM layer
- Solution I
- StoRM with GPFS
- Results
- Solution II
- StoRM with GPFS and TSM
- Results
- Conclusions
16StoRM with GPFS and TSM
- We combined GPFS specific features (available
commencing version 3.2) and TSM (also by IBM)
with StoRM to provide a transparent Grid-enabled
Hierarchical Storage Manager (HSM) solution - An interface between GPFS and TSM has been
realized in order to allow for intelligent tape
recalls. By means of this interface, files are
recalled in the best order as they appear on
tape. - StoRM has been extended to include the SRM
methods required to manage the tapes - Pre-production test-bed built to satisfy the
scale of our largest experiment - First stress tests on GPFS-TSM only
- Verification of the GPFS-TSM integration and
scalability - User tests on complete system
17HW set-up
8 tape drives T10KB - 1 TB per tape, - 1 Gbps
per drive
18GPFS with TSM validation tests
- Concurrent r/w accesses to the storage for tape
migrations and recalls, and from batch nodes
where analysis jobs run - StoRM not used in these tests
- 3 HSM nodes serving 8 T10KB drives
- 6 drives (at maximum) used for recalls
- 2 drives (at maximum) used for migrations
- Order of 1GB/s of aggregated traffic
- 550 MB/s from tape to disk
- 100 MB/s from disk to tape
- 400 MB/s from disk to the computing nodes (not
shown in this graph)
19Results with StoRM
- 24 TB of data moved from tape to disk
- Recalls of five days typical usage by a large LHC
experiment (namely CMS) compacted in one shot and
completed in 19h - Files were spread on 100 tapes
- Average throughput 400MB/s
- 0 failures
- Up to 6 drives used for recalls
- Simultaneously, up to 3 drives used for
migrations of new data files
400 MB/s
20Overview
- INFN Tier1
- Mass Storage Challenge
- SRM layer
- Solution I
- StoRM with GPFS
- Results
- Solution II
- StoRM with GPFS and TSM
- Results
- Conclusions
21Conclusions
- We implemented a full HSM system based on GPFS
and TSM able to satisfy the requirements of a
WLCG Tier-1 centre like INFN-CNAF - StoRM, the Storage Resource Manager software
layer developed at CNAF, has been extended in
order to cope with tape support. - We tested the overall system (StoRM/GPFS/TSM)
with excellent performance results - We were able to fully saturate the bandwidth of
the 8 tape drives in use (1 Gbps each) reaching
about 800 MB/s of aggregated throughput to/from
the tape drives - We migrated our largest HSM user (the LHC
experiment CMS) to StoRM and more will follow in
the near future