5 Ways to Get Better
Performance from
NAS-backed VMs
Best practices and emerging technologies
Welcome!
• Originally presented by
Peter Smith, Director of Product Management
– @PeterSmith2013
• On-demand webinar avai...
Agenda
• Best practices for VMware on NAS
– Host
– Network
– Storage
• Emerging performance technologies
– 10Gb Ethernet
...
Best practices for VMware on NAS
• Host
– SIOC, NIOC, DRS, VAAI
• Network
– Load sharing, link aggregation,
jumbo frames
...
Host optimization
• SIOC—Storage I/O Control
– Workload throttling based on
storage latency thresholds
• NIOC—Network I/O...
Host optimization
• VAAI—VMware vSphere Storage APIs
Array Integration
– APIs between VMware and storage
– Keeps storage o...
Network optimization
• Multiple datastores with separate
NFS connections
– Break datastores over multiple
TCP connections
...
Network optimization
• Link Aggregation
– Stitch together multiple network links
into one big pipe
• Jumbo frames
– Maxim...
Storage optimization
• Sizing
– Volume size
– Block size
– Rsize and Wsize
– Max VMs per NFS
• TCP/IP heap size
• TCP co...
Storage optimization
• Maximum number of NFSs per ESXi
• Thin provisioning
• Deduplication
Emerging performance technologies
• 10Gb Ethernet
• SR-IOV
• VMDq
• Edge caching
10Gb Ethernet
• Much better throughput
– Fatter pipe, but can the storage handle the traffic?
• Easier management
– Fewer...
Single Root IO Virtualization (SR-IOV)
• Removes latency caused by the
software stack
– HDDs: sw = 1% of latency
– SSDs: s...
Virtual Machine Device Queues (VMDq)
• Chip-based technology
– “A LAN device that provides queues
for software-based NIC s...
Edge caching
• Brings the content closer to the demand
– Data at rest still lives on spinning disk
– Active data is brough...
Edge caching
• Two approaches:
– SSD-based (HW+SW)
– Memory-based (SW only)
• Provides biggest overall improvement
– Mino...
How edge caching stacks up
10GbE
SR-IOV
VMDq
Edge caching
Manageability
—
Complex
—
—
CPU affect
—
—
Offloads
...
Edge caching: HW/SW vs SW-only
HW/SW
• Local only
Software-only
• Distributed
• Only affects attached host
• Pooled amo...
Infinio Accelerator
A VMware environment supported by NFS storage
Infinio Accelerator
VMware/NAS environment with Infinio deployed
Infinio Accelerator Dashboard
Infinio Accelerator Dashboard
Thank you!
www.infinio.com
@InfinioSystems
@PeterSmith2013
On-demand webinar available
(BrightTalk login required)
of 23

5 Ways to Get Better Performance from NAS-backed VMs

Based on current practices in the datacenter and key emerging technologies, now is the time to revisit the debate on performance in NAS-backed virtual environments. This presentation covers recommended practices and configuration tips, and shows how state-of-the-art technology changes the cost/performance dynamic for datacenter operations. >Learn recommended practices and configuration tips to improve application performance and scalability of NAS-backed virtual environments. >Assess ways to optimize the host, network, and storage tiers including jumbo frames and link aggregation performance, congestion control and prioritization mechanisms, and common configuration pitfalls. >Evaluate four trending technologies, including 10Gb Ethernet, SR-IOV, VMDq, and edge caching. Evaluate each along the dimensions of performance (throughput, latency) and ease of use, with a cost/benefit summary to illustrate which offers the greatest performance gain for the dollar.
Published on: Mar 3, 2016
Published in: Technology      
Source: www.slideshare.net


Transcripts - 5 Ways to Get Better Performance from NAS-backed VMs

  • 1. 5 Ways to Get Better Performance from NAS-backed VMs Best practices and emerging technologies
  • 2. Welcome! • Originally presented by Peter Smith, Director of Product Management – @PeterSmith2013 • On-demand webinar available (BrightTalk login required) • Tweet us questions: @InfinioSystems
  • 3. Agenda • Best practices for VMware on NAS – Host – Network – Storage • Emerging performance technologies – 10Gb Ethernet – SR-IOV – VMDq – Edge caching
  • 4. Best practices for VMware on NAS • Host – SIOC, NIOC, DRS, VAAI • Network – Load sharing, link aggregation, jumbo frames • Storage – Sizing, NFS/ESXi, TCP settings, thin provisioning, deduplication
  • 5. Host optimization • SIOC—Storage I/O Control – Workload throttling based on storage latency thresholds • NIOC—Network I/O Control – Rule-based allocation of bandwidth to different traffic types • DRS—Distributed Resource Scheduler – Load balancing of server resources across hosts
  • 6. Host optimization • VAAI—VMware vSphere Storage APIs Array Integration – APIs between VMware and storage – Keeps storage ops and management in storage system • Storage DRS – Dynamic balancing of workloads across datastores
  • 7. Network optimization • Multiple datastores with separate NFS connections – Break datastores over multiple TCP connections • Load sharing (subnets, IP hash) – Separate subnets across multiple network links – Each single connection limited to one link’s bandwidth
  • 8. Network optimization • Link Aggregation – Stitch together multiple network links into one big pipe • Jumbo frames – Maximize throughput – Align frames size with workload element size – Make bandwidth use more efficient
  • 9. Storage optimization • Sizing – Volume size – Block size – Rsize and Wsize – Max VMs per NFS • TCP/IP heap size • TCP connections per IP
  • 10. Storage optimization • Maximum number of NFSs per ESXi • Thin provisioning • Deduplication
  • 11. Emerging performance technologies • 10Gb Ethernet • SR-IOV • VMDq • Edge caching
  • 12. 10Gb Ethernet • Much better throughput – Fatter pipe, but can the storage handle the traffic? • Easier management – Fewer NICs • Only minimal latency improvement – Tens of microsecond faster
  • 13. Single Root IO Virtualization (SR-IOV) • Removes latency caused by the software stack – HDDs: sw = 1% of latency – SSDs: sw = 20% of latency • Virtualized IO – Virtualized functions available to each VM – Bypasses software layers in the data path • Management of SR-IOV is complex • Performance improvement in microseconds
  • 14. Virtual Machine Device Queues (VMDq) • Chip-based technology – “A LAN device that provides queues for software-based NIC sharing” –Intel • Offloads IO management from CPU – As VMs and traffic increase more CPU is used, leaving less for apps/VMs – VMDq offloads IO burden, provides management & switching
  • 15. Edge caching • Brings the content closer to the demand – Data at rest still lives on spinning disk – Active data is brought closer to where it's needed – Improvements in milliseconds, not microseconds
  • 16. Edge caching • Two approaches: – SSD-based (HW+SW) – Memory-based (SW only) • Provides biggest overall improvement – Minor consumption of underutilized host resources – Major performance increase from offloaded requests
  • 17. How edge caching stacks up 10GbE SR-IOV VMDq Edge caching Manageability — Complex — — CPU affect — — Offloads None/minor Platform HW (network) HW (PCI) HW (silicon) HW (SSD) or SW Throughput Increased — — Offloaded Latency improvement Microseconds Microseconds Microseconds Milliseconds (1/1,000,000 sec) (1/1,000,000 sec) (1/1,000,000 sec) (1/1,000 sec)
  • 18. Edge caching: HW/SW vs SW-only HW/SW • Local only Software-only • Distributed • Only affects attached host • Pooled among hosts • Requires hardware at every server • No hardware to install – Installation, reboots… – Takes advantage of underutilized resources • Can’t be shared—cache sprawl • Shareable—better efficiency • Popular in FC, not much interest in NFS • Ease-of-use
  • 19. Infinio Accelerator A VMware environment supported by NFS storage
  • 20. Infinio Accelerator VMware/NAS environment with Infinio deployed
  • 21. Infinio Accelerator Dashboard
  • 22. Infinio Accelerator Dashboard
  • 23. Thank you! www.infinio.com @InfinioSystems @PeterSmith2013 On-demand webinar available (BrightTalk login required)

Related Documents