VSWAPPER A MEMORY SWAPPER FOR VIRTUALIZED ENVIRONMENTS 1

VSWAPPER: A MEMORY SWAPPER FOR VIRTUALIZED ENVIRONMENTS 1 NADAV AMIT, DAN TSAFRIR, ASSAF SCHUSTER

VIRTUALIZATION • Hardware virtualization enables the cloud ecosystem • Higher server utilization • Up to 50% less operating expenses • Up to 80% less power expenses • Memory is the biggest constraint for consolidation [IDC’ 09, Birke’ 12] • Essential for server consolidation App. VM Hypervisor 2 • Memory overcommitment

P 1 RAM disk B RAM host view P 2 (1) Access P 1 P 3 P 2 Guest Code (2) Exit disk VM image host swap area P 2 (3) Swap Hypervisor Poor performance due to the semantic gap 3 guest view UNCOOPERATIVE SWAPPING

MEMORY BALLOON inflate guest memory deflate guest memory balloon swap out virtual disk swap in • Allows guest to make paging decisions, yet: • Virtualization vendor use host swapping as a backup 4 • Ballooning takes time • Not a complete solution • Paravirtual

DEMONSTRATION: SEQUENTIAL FILE READ 512 MB VM; 100 MB cap; 200 MB file; Read & re-read Swapping st Disk read Balloon 5 ac c Files e n gu i d e h y; mor e m t s ho not in

AGNEDA • Introduction • Problems • Solutions 6 • Evaluation

PROBLEM #1: STALE SWAP READS 7 Stale swap reads

guest view PROBLEM #1: STALE SWAP READS RAM P 2 VM reads P 2 disk P 2 (2) RAM host view (1) P disk VM image host swap area P 2 P #PF reading P P P 2 (3) overwrite it with P 2 P 2 P 8 Swap data is read just for disk blocks to overwrite it

PROBLEM #2: FALSE SWAP READS (1) host view RAM disk VM image host swap area VM writes to P P (2) P VM overwrites P P Guest reallocates page frames: • Copy-on-write • Slab pages 9 • Zero page before use

PROBLEM #3: SILENT SWAP WRITES host view disk VM image host swap area P P RAM P (2) (1) vm reads page P host swaps page P P 10 Data read from the image is written back to the host swap

11 PROBLEM #4: DECAYED SWAP SEQUENTIALITY

PROBLEM #4: DECAYED SWAP SEQUENTIALITY host view RAM disk VM image host swap area P 1 P 2 P 3 P 1 VM image data gradually loses sequentiality on swap 12 Poor prefetch

PROBLEM #5: FALSE ANONYMITY anon host view RAM Userspace hypervisor code guest: named host: anon P 4 P 1 P 3 QEMU disk VM image P 1 Swap out P 2 P 3 host swap area OS prefers to evict named pages; hypervisor ignores 13 All guest memory is considered anonymous

14 DEMONSTRATION: SEQUENTIAL FILE READ

AGNEDA • Introduction • Problems • Solutions 15 • Evaluation

SOLUTIONS • Extension to existing swapping • Full-virtualization • No VM introspection • Based on general OS concepts • Can be used with ballooning • Two mechanisms 16 • Swap mapper • False read preventer

P RAM (1) Read command disk B (2) Exit host view RAM P disk VM image B (3) File read Associate guest memory page with disk block (P=B) (3) mmap to file (4) Remove mapping on write to P or B Similar flow for writes 17 guest view SOLUTION: SWAP MAPPER

SOLUTION: SWAP MAPPER Solves: • • • Subtleties: • • • Stale swap reads – map instead of read Silent swap writes – no WB of clean pages Decayed swap sequentiality – Data in its original place False anonymity – pages from image are named Consistency VM disk uses 4 KB blocks Prefetch only if present in guest physical memory Remove from host cache if overwritten Limitations • • Overheads – CPU and memory Cannot track migrated pages 18 •

SOLUTION: FALSE READ PREVENTER mov [0 x 10], 5 • Save VM writes to buffer Virtual Machine Exit • If not, fetch asynchronously and merge Hypervisor • Subtleties: • Selective emulation • Efficient emulation • Host reads Entry Fetch Record Map [0 x 10]=5 Writeback …. 5…. 19 • If page is rewritten, allocate one from free pool and remap

AGNEDA • Introduction • Problems • Solutions 20 • Evaluation

EVALUATION IBM R 420 Server KVM Hypervisor Enterprise disk baseline: host-swapping, no host caching mapper: swap mapper vswapper: swap mapper and false read preventer balloon: paravirtual balloon, host swapping enabled 21 Configurations

PBZIP 2 VM: 512 MB, 1 VCPU 1. 63 x speedup 22 OOM

KERNBENCH 23 VM: 512 MB, 1 VCPU

DYNAMIC WORKLOAD: METIS MAP-REDUCE (WC) VM: 2 GB; Host: 8 GB Mo. M manages balloon VMs started 10 sec. apart 24 2 x speedup (w/balloon)

RELATED WORKS • Monitoring the buffer-cache [Jones 06] • Non-blocking writes [Useche 12] • Memory overcommit mechanisms 25 • Transcendent memory [Magenheimer’ 09] • Cooperative memory management [Schwidefsky’ 06] • Memory hot-plug [Schopp’ 06]

CONCLUSION • Host-swapping poor performance analysis and solution • Swap mapper • False read preventer • VSwapper integrates with balloon 26 • Vswapper code is available nadav. amit. to/vswapper

27 QUESTIONS