1 / 16

INFN-GRID-WP4: Computing Fabric & Mass Storage Attivita’ 2003

INFN-GRID-WP4: Computing Fabric & Mass Storage Attivita’ 2003 ( CNAF, Genova, Lecce, LNL, Padova, Napoli, Roma1). WP Breakdown. Fabric Design

thane
Download Presentation

INFN-GRID-WP4: Computing Fabric & Mass Storage Attivita’ 2003

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. INFN-GRID-WP4: Computing Fabric & Mass Storage Attivita’ 2003 (CNAF, Genova, Lecce, LNL, Padova, Napoli, Roma1)

  2. WP Breakdown Fabric Design • l’attivita’ del 2003 e’ focalizzata sulle tecniche di storage di una farm con un numero elevato di nodi. L’utilizzo di NFS in questi casi e’ di difficile realizzazione e fornisce prestazioni modeste. Ci sono vari approcci possibilie. Il piu interessante per noi e’ quello legato alla possibilita’ di SAN a basso costo associate ad un file system condiviso da tutta la farm. Quindi l’attivita’ del 2003 e’ cosi articolata • Commodity Storage Area Network (SAN) • Cluster file systems Fabric Management • L’attivita 2003 continua il programma concordato con wp4-edg. • Installation Task • Transizione LCFG->LCFGng->EDG wp4 tools

  3. Fabric Management: Installation Task The INFN efforts in WP4 has been spent during the review period as follows: • INFN Legnaro (E. Ferro and C. Aifitimiei): • LCFGng deployment and support: maintenance of the existing documentation for Red Hat 6.2 server, add of the installation guide for Red Hat 7.3 (thanks to WP6 collaboration), support for deployment on the testbed (bug fixes and some improvements). 1 PM. • [WP4I-16] Automated installation infrastructure: Design. Almost complete, announced to mailing lists, waiting for comments. 0.6 PM. • [WP4I-17] Automated installation infrastructure: installation server. Started implementation only of basic functionalities (DHCP configuration). 0.2 PM • INFN CNAF (A. Chierici ): • Generic support to CNAF-Tier1 staff for the installation problem with LCFG. • Feedback to WP4/WP6 on the deployment of the new LCFGng installation system. • Development, in conjunction with WP1 staff, of 2 new LCFGng objects: jssrbconfig and uiconfig.

  4. Fabric Design • L’attivita del 2003 e’ si compone di due argomenti strettamente correlati tra loro: • commodity SAN (o presunte) • gigaethernet (con migrazione a 10 Geth se possibile) • Infiniband • Cluster File Systems da poter utilizzare con una commodity SAN e dove quindi e’ immaginabile un file system condiviso tra tutti i nodi di calcolo e disk server • Partecipanti: • Genova • LNL • Napoli

  5. Application Servers Data Servers NFS gateway Server S A N Disk Array Server Disk Array Disk Array Server Modello di storage con NFS • - Attualmente l’accesso remoto ai Data Server • e’ penalizzato da NFS • Dato l’elevato numero di Application Server • ci sono scarsi vantaggi ad utilizzare SAN basate • per esempio su FC (costoso)

  6. Commodity SAN Application Servers Data Servers GEth/iSCSI SAN enabled disk array Infiniband gateway SAN low cost (commodity) permettono di raggiungere direttamente tutti i nodi. Di interesse per noi sono: GigaEthernet con trasporto SCSI su IP (iSCSI) Infiniband L’integrazione completa sulla farm richiede cmnq un file system a livello di farm stessa.

  7. Intel GE 1000 T IP Storage Adapter iSCSI • Internet SCSI (iSCSI) is a standard protocol for encapsulating SCSI command into TCP/IP packets and enabling I/O block data transport over IP networks • iSCSI adapters combines NIC and HBA functions. • take the data in block form • handle the segmentation and processing with TCP/IP processing engine • send IP packets across the IP network Storage HBA FC Storage iSCSI Adapter Network Interface Card Application Layer File Block Block IP Server FC Server IP Server Block Block Driver Layer IP Packets Link Layer IP Packets IP Packets FC Packets on Ethernet on Ethernet

  8. iSCSI products Application Servers GEth/iSCSI Data Servers LSI iMegaRAID 1 = ~ 5 Tbyte x controller SATA 2 x GE 16 • Host adapter: • Intel GE 1000 T • Adaptec ASA-7211 • LSI 5201 • ecc. iSCSI Controller RAID – SATA Controller

  9. Infiniband Transport Protocols • IBA has been developed with Virtual Interface in mind. VIPL 2.0 includes IBA extensions and RDMA operations. • SCSI RDMA Protocol (SRP). It is a T10 standard. • SRP defines mapping to IBA architecture • it is the transport protocol over IBA • SRP is based on VI • Direct Access Files System (DAFS) • Direct Access Socket (DAS) • TCP/IP functionality over VI/IB fast, low latency network storage fast, low latency TCP/IP sockets File Access Block Access SRP DAFS DAS Virtual Interface over IB IBA Host Channel Adapter

  10. LNL Infiniband Test Bed • All the hardware has been provided by Infiniswitch (1 switch + 4 HCA) • All the hardware is up and running • First p2p measurements have been performed • Software • Virtual Interface Library (VIPL) as provided by Infiniswitch • Send/Receive over reliable connections • RDMA over reliable connections • Sourceforge has a infiniband project over Linux • VIPL source is available. Compiled and works! • Performance as the infiniswitch VIPL (probably they are the same code) • Results • Round trip time small buffers ~ 40 msec (latency 20 msec) • P2P ~ 80% link saturation

  11. LNL Infiniband Test Bed II Leaf Switch 32 1x (2.5 Gbps) ports in 1 U chassis IBA Host Channel Adapter PCI-X (max ~ 380 Mbyte/s) Supermicro P4DL6

  12. Infiniband: risultati ottenuti Link Saturation 220

  13. Cluster File Systems Application Servers Data Servers Cluster File System • GFS • GPFS • Lustre • Storage Tank

  14. Legnaro Test Bed IBM Blade center Read 30 Mbyte/s Write 30 Mbyte/s • Blade + 4 server + 200 GB*4 • Test con GPFS • Test in programmazione fino a giugno: • valutazione lustre – storage tank (cluster file system) su FC o su DAS se possibile • Test bed su iscsi con : • Host adapter LSI 5201 • iMegaRAID solution GPFS DAS Disk Servers

  15. Napoli test bed • La proposta e’ di equipaggiare la farm di FIRB (16 macchine) con una rete Myrinet a 2 Gbps • Durante il setup di FIRB la farm viene utilizzata anche per fare I test sui cluster file system • A test finiti Myrinet rimane a FIRB abilitando la farm al calcolo parallelo (con MPI) ad alta velocita. • Myrinet, essendo l’unico link per cluster commerciale a 2 Gbps, permette il test dei cluster file system ad alta velocita’.

  16. Genova test bed • Viene potenziata la loro Farm Gethernet aumentando il numero di nodi (+ 15 PC) • L’attivita’ continua su: • Protocollo Gamma • in produzione su applicativi paralleli, portato su MPI • DRAID • Progetto completato • Test di scalabilita’

More Related