difference engine:  harnessing memory redundancy in virtual machines by diwaker gupta et al

Post on 03-Jan-2016

21 Views

Category:

Documents

2 Downloads

Preview:

Click to see full reader

DESCRIPTION

Difference Engine:  Harnessing Memory Redundancy in Virtual Machines by Diwaker Gupta et al. presented by Jonathan Berkhahn. Motivation. Virtualization has improved and spread over the past decade Servers often run at 5-10% of CPU Capacity High capacity needed for peak workloads - PowerPoint PPT Presentation

TRANSCRIPT

Difference Engine: Harnessing Memory Redundancy

in Virtual Machinesby Diwaker Gupta et al.

 

presented by Jonathan Berkhahn

Motivation

• Virtualization has improved and spread over the past decade

• Servers often run at 5-10% of CPU Capacityo High capacity needed for peak workloadso Fault isolation for certain serviceso Certain services run best on particular configurations

  • Solution: Virtual Machines

 

Problem

• CPU's suited to multiplexing, main memory is not• Upgrading not an ideal option

o Expensiveo Limited by slots on the motherboardo Limited by ability to support higher capacity moduleso Consumes significant power, and therefore produces

significant heat• Further exacerbated by current trends toward many-core

systems

 

   How do we fix this memory bottleneck for virtual machines?

Difference Engine

• Implemented as an extension to the Xen VMM o Sub-page granularity page sharingo In-memory page compression

 • Reduces the memory footprint by up to 90% for

homogenous workloads and up to 65% for heterogeneous workloads

Outline

• Related Work• Difference Engine algorithms• Implementation• Evaluation

Page Sharing

• Transparent page sharingo Requires guest OS modification

 • Content-based

o VMWare ESX

Delta Encoding

• Manbero Rabin fingerprintso Inefficient

• Brodero Combined Rabin fingerprints and sampling

•  Both focused on identifying similar files, but not encoding the differences

Memory Compression

• Douglis et al.o Sprite OSo Double-edge sword

• Wilson et al.o Previous results due to slow hardwareo Developed algorithms that exploit virtual memory

structure

Outline

• Related Work• Difference Engine algorithms• Implementation• Evaluation

Page Sharing

• Content-based• Hash pages and index by hash value

o Hash collisions indicate a potential matcho Compare byte-by-byte to ensure pages are

identicalo Reclaim one page, update virtual memoryo Writes cause a page fault trapped by the VMM

Patching

• Sharing of similar pages• Identify similar pages, store differences as a

"patch"• Compresses multiple pages down to single

reference copy and a collection of patches

Identifying Candidate Pages

 

Compression

• Compression of live pages in main memoryo Useful only for high compression ratios

• VMM traps requests for compressed pages

Overview

 

Paging Machine Memory

• Last resort • Copy pages to disk• Extremely expensive operation• Leaves policy decisions to end user

Caveat

Both patching and compression are only useful for infrequently accessed pages.

   

So, how do we determine "infrequent"?

Clock

• Not-Recently Used policy• Checks if page has been referenced/modified

o C1 - Recently Modifiedo C2 - Recently Referencedo C3 - Not Recently Accessedo C4 - Not Accessed for a While

Outline

• Related Work• Difference Engine algorithms• Implementation• Evaluation

Implementation

• Modification to Xen VMM• Roughly 14,500 lines of code, plus 20,000 for ports of

existing patching and compression algorithms• Shadow Page Table

o Difference Engine relies on modifying the shadow page and P2M tables

o Ignored pages mapped by Dom-0 • Complications: Real Mode and I/O support

Complications

• Booting on bare metal disables pagingo Requires paging to be enabled within guest OS

• I/Oo Xen hypervisor emulates I/O hardware with a Dom-0

process ioemu, which directly accesses guest pageso Conflicts with policy of not acting on Dom-0 pageso Unmap VM pages every 10 seconds

Clock

• NRU policy• Tracked by Referenced and Modified bits on each

page• Modified Xen's shadow page tables to set bits

when creating mappings• C1 - C4

Page Sharing

• Hash table in Xen heapo Memory limitations - 12 Mb

• Hash table only holds entries for 1/5 memoryo 1.76 Mb hash table

• Covers all of memory in 5 passes

Detecting Similar Pages

• Hash Similarity Detector (2,1) o Hash similarity table cleared after all pages have

been considered• Only building the patch and replaced the page

requires a locko May result in a differently sized patch, but will

still be correct

Compression & Disk Paging

• Antagonistic relationship with patchingo Compressed/Disk pages can't be patched

• Delayed until all pages have been checked for similarity and the page has not been accessed for a while (C4) 

 • Disk paging done by daemon running in Dom-0

Disk Paging

 

Outline

• Related Work• Difference Engine algorithms• Implementation• Evaluation

Evaluation

• Experiments run on dual-processor, dual-core 2.33 GHz Intel Xeon, 4 KB page size

• Tested each operation individually for overhead

Page Lifetime

 

Homogenous VMs

 

Homogenous Workload

 

Heterogeneous Workload

 

Heterogeneous Workload 2

 

Utilizing Savings

 

Conclusion

• Main memory is a primary bottleneck for VMs• Significant memory savings can be achieved from:

o Sharing identical pageso Patching similar pageso In-memory page compression

• Implemented DE and showed memory savings of as much as 90%

• Saved memory can be used to run more VMs

 

Discussion

top related