We present the design, implementation, and evaluation of post-copy based live migration for virtual machines (VMs) across a Gigabit LAN. Live migration is an indispensable feature in today's virtualization technologies. Post-copy migration defers the transfer of a VM's memory contents until after its processor state has been sent to the target host. This deferral is in contrast to the traditional pre-copy approach, which first copies the memory state over multiple iterations followed by a final transfer of the processor state. The post-copy strategy can provide a "win-win" by reducing total migration time closer to its equivalent time achieved by non-live VM migration. This is done while maintaining the liveness benefits of the pre-copy approach. We compare post-copy extensively against the traditional pre-copy approach on top of the Xen Hypervisor. Using a range of VM workloads we show improvements in several migration metrics including pages transferred, total migration time and network overhead. We facilitate the use of post-copy with adaptive pre-paging in order to eliminate all duplicate page transmissions. Our implementation is able to reduce the number of network-bound page faults to within 21% of the VM's working set for large workloads. Finally, we eliminate the transfer of free memory pages in both migration schemes through a dynamic self-ballooning (DSB) mechanism. DSB periodically releases free pages in a guest VM back to the hypervisor and significantly speeds up migration with negligible performance degradation.

/pdf/post-copy-based-live-virtual-machine-migration-using-16z7hbldrv.pdf

Post-copy based live virtual machine migration using adaptive pre-paging and dynamic self-ballooning

We present the design, implementation, and evaluation of post-copy based live migration for virtual machines (VMs) across a Gigabit LAN. Post-copy migration defers the transfer of a VM's memory contents until after its processor state has been sent to the target host. This deferral is in contrast to the traditional pre-copy approach, which first copies the memory state over multiple iterations followed by a final transfer of the processor state. The post-copy strategy can provide a "win-win" by reducing total migration time while maintaining the liveness of the VM during migration. We compare post-copy extensively against the traditional pre-copy approach on the Xen Hypervisor. Using a range of VM workloads we show that post-copy improves several metrics including pages transferred, total migration time, and network overhead. We facilitate the use of post-copy with adaptive prepaging techniques to minimize the number of page faults across the network. We propose different prepaging strategies and quantitatively compare their effectiveness in reducing network-bound page faults. Finally, we eliminate the transfer of free memory pages in both pre-copy and post-copy through a dynamic self-ballooning (DSB) mechanism. DSB periodically reclaims free pages from a VM and significantly speeds up migration with negligible performance impact on VM workload.

/pdf/post-copy-live-migration-of-virtual-machines-1r78fdaulv.pdf

Post-copy live migration of virtual machines

Numerical transient analysis of Markov models

Memory transfers due to a cache miss are costly. Prefetching all memory references in very fast computers can increase the effective CPU speed by 10 to 25 percent.

Sequential Program Prefetching in Memory Hierarchies

Several different models for predicting coverage in a fault-tolerant system, including models for permanent, intermittent, and transient errors, are discussed. Markov, semi-Markov, nonhomogeneous Markov, and extended stochastic Petri net models for computing coverage are developed. Two types of events that interfere with recovery are examined; and methods for modeling such events, whether they are deterministic or random, are given. The sensitivity of system reliability/availability to the coverage parameter and the sensitivity of the coverage parameter to various error-handling strategies are investigated. It is found that a policy of attempting transient recovery upon detection of an error (as opposed to automatically reconfiguring the affected component out of the system) can actually increase the unreliability of the system. >

Coverage modeling for dependability analysis of fault-tolerant systems

In this paper, on-line algorithms for division and multiplication are developed. It is assumed that the operands as well as the result flow through the arithmetic unit in a digit-by-digit, most significant digit first fashion. The use of a redundant digit set, at least for the digits of the result, is required.

On-Line Algorithms for Division and Multiplication

A review and a critical evaluation of a representative class of state-of-the-art models for ultrahigh reliability prediction is presented. This evaluation naturally leads us to a new model for ultrahigh reliability prediction now under development. The new model combines the flexibility and accuracy of simulation with the speed of analytic models.

Ultrahigh Reliability Prediction for Fault-Tolerant Computer Systems

In order to remain tractable, mhany reliability models do not include the states and transitions necessary to represent fault/error-handling details. Instead, the effectiveness of fault/ error-handling mechanisms is represented by the use of instantaneous coverage probabilities. This paper investigates the effect of the error introduced by the assumption of instantaneous coverage probabilities on the predictions of the reliability model, and it shows that the reliability estimates thus obtained are lower bounds on the reliability estimates of the composite model with embedded fault/error-handling states and transitions. The paper also discusses the choice of the calculation method for the instantaneous coverage probabilities and defines a near-coincident-fault coverage model that yields conservative instantaneous coverage probabilities.

The Conservativeness of Reliability Estimates Based on Instantaneous Coverage

Data paging is of primary concern for problems with large data bases and for many types of array problems. We show that prepaging reduces the paging problems of array algorithms operating on large arrays. We also show that the use of a submatrix algorithm considerably improves the locality. Finally, we consider methods of automating these performance-improvement techniques by means of a compiler in the context of a structured array language.

On the Paging Performance of Array Algorithms

A demand prepaging algorithm DPMIN is defined and proved to be an optimal demand prepaging algorithm. However, it cannot be used in practice since it requires that the future refreence string be completely known in advance. Several practical prepaging algorithms are also defined which require only a partial knowledge of the future reference string. Finally, we show that these prepaging algorithms reduce the paging problems of array algorithms operating on large arrays.

Trivedi

Papers

On-Line Algorithms for Division and Multiplication

Ultrahigh Reliability Prediction for Fault-Tolerant Computer Systems

The Conservativeness of Reliability Estimates Based on Instantaneous Coverage

On the Paging Performance of Array Algorithms

Prepaging and Applications to Array Algorithms