1 / 12

Modeling and Performance Evaluation of iSCSI SANs over TCP/IP MANs and WANs

Modeling and Performance Evaluation of iSCSI SANs over TCP/IP MANs and WANs. C. M. Gauger, M. Köhn, S. Gunreben, D. Sass, S. G. Perez Institute of Communication Networks and Computer Engineering Universität Stuttgart {gauger, koehn, gunreben, sass}@ikr.uni-stuttgart.de. Objectives and Motivation.

danil
Download Presentation

Modeling and Performance Evaluation of iSCSI SANs over TCP/IP MANs and WANs

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. Modeling and Performance Evaluation of iSCSI SANs over TCP/IP MANs and WANs C. M. Gauger, M. Köhn, S. Gunreben, D. Sass, S. G. PerezInstitute of Communication Networks and Computer EngineeringUniversität Stuttgart{gauger, koehn, gunreben, sass}@ikr.uni-stuttgart.de

  2. Objectives and Motivation • SANs gain importance as network application due to • Storage consolidation, IT virtualization • Business continuity and disaster recovery • Terrorist attacks, post 9/11 regulations • Natural catastrophes, c.f. earthquake risc in Bay Area SAN extensions over larger distances (approx. > 200km) • SANs evolve from dedicated equipment and networks to standard equipment and network services • Architectures • FiberChannel (FC): mostly local, expensive equipment, own addressing, own isolated high-QoS networks • FCIP, iFCP: FC extensions over IP networks (tunneling) • iSCSI: native IP protocols, addressing, TCP/IP networks Q: Requirements towards MAN/WAN networks?  Impact of distance and network QoS on iSCSI performance

  3. iSCSI SAN extension scenario Server with local primary storage Secondary storage TCP/IP network, e.g. enterprise VPN iSCSI initiator iSCSI target SCSI requests TCP/IP network, e.g. enterprise VPN

  4. iSCSI • Standardized by IETF in RFC 3720, 3980 • Lightweight protocol • Encapsulates and transports SCSI requests • Relies on underlying TCP/IP network for failure-free transport • Simple block-based flow control • To be implemented in software • Hosts most likely to be in Ethernet (GbEth) LANs • Inexpensive NICs • NIC can employ TCP-offload engine (TOE) to speed-up TCP processing and offload hosts

  5. Modeling and Evaluation scenario • Traffic models • Empirical I/O traffic model (IBM) • Modified empirical and generic I/O traffic models • Network model • Packet loss probability (10-2, 10-3, 10-6) • Round-trip time (initiator-target distance: 0...1000km) • TCP/IP network abstraction model • Bandwidth determined by • Packet loss probability • RTT • TCP window sizes • Access bandwidth • Models by Padhye et al. and Mathis et al. • All details, c.f. report

  6. Modeling • Storage service: asynchronous write across n/w • iSCSI scenarios • Single, isolated request Twrite=f(RTT, p, S, Tproc, ...) • Dynamic request traffic without interleaving • Dynamic request traffic with interleaving

  7. Impact of RTT and QoS • Throughput limited by TCP: both RTT and losses • Short requests, throughput limited by iSCSI • Larger requests • Given RTT: better QoS can even double throughput • Given throughput: better QoS can increase distance significantly • QoS: few differences below p=10-4

  8. Impct of processing delay • Decreasing Tproc increases throughput • For small requests up to factor of 4 • For large requests smaller gains • Few benefit for lossy networks

  9. Results for dynamic traffic – no interleaving • Under dynamic traffic queueing occurs • Queueing delay increase with RTT (c.f. Twrite) • Significant additional delay strongly depending on traffic model • For loss <10-3 few impact of QoS

  10. Results for dynamic traffic – with interleaving • With interleaving of requests in an iSCSI session transfer of several requests can be parallelized • Queueing delay significantly reduced with Twrite as RTT now only limits TCP but not iSCSI any more • For large requests, QoS becomes dominant for given RTT

  11. Impact of offered load • Service time of the request queueing system is not only determined by the transmission time of requests but by processing and transfer overheadQ: when is stability limit reached for the request queues? • Great improvement with interleaving • Strong impact of processing delay • Remaining link capacity can be used by parallel iSCSI session Load=1 with respect to 1Gbps link Tproc=0ms with interleaving shows max. TCP throughput for this scenario

  12. Conclusions • Modelling and performance evaluation of principal iSCSI behavior over MAN/WAN networks • Single request model • Throughput limited by TCP: both RTT and losses • QoS: few differences below p=10-4 • Dynamic request traffic • Significant impact of traffic model • Interleaving greatly improves performance for larger distances • QoS more dominant for larger requests with interleaving • Detailed models and results available in report (inline object below) or on NOBEL server, WP2 Madrid, also submitted for publication

More Related