Abstract: Managing resources in a VM cluster; allocation of resources among competing VMs. Assigning VMs to real devices, responsive to needs for resources: processor usage, disk I/O, network I/O, memory space, disk space. Assigning VMs responsive to needs for cluster activity: network response latency, QoS. Using predictive models of resource usage by VMs. Transferring VMs, improving utilization. Failing-soft onto alternative VM resources. Providing a resource buffer for collective VM resource demand. Transferring VMs to a cloud service that charges for resources.

Abstract: Providing differing degrees of reliability or security for distinct devices within a reliable distributed system. Allowing virtual machines to operate in a reliable distributed system without either exposing the actual available devices to guest operating systems, or requiring the guest OS to maintain, or to maintain information about, reliability or security. Methods are responsive to a hierarchy of possible reliability or security failures, to guard more carefully against simultaneous failures of reliability or breaches of security, without additional work or delay. Functions invoked by the user that relate to reliability and security are hidden, so the user can specify a degree of reliability or security without having to implement those requirements themselves. Providing a uniform set of resources available to all users, separating out those individual resources that are allocated to particular users, and emulating particular devices at the request of those particular users.

Abstract: In a reliable distributed computing system, techniques where user actions or events affecting real system devices (RSD's), or causing the system to make changes to metadata, require fewer accesses to RSD's. When a virtual system device (VSD) is mounted, substantially all the metadata, or at least that associated with data recovery, is read into memory. Changes to that metadata can be made substantially without any metadata accesses to RSD's. No metadata accesses to RSD's are made for read operations, as the RSD would remain unchanged. Metadata is flushed from memory upon write operations. Metadata about data locations can be maintained in memory, so as to be available if any RSD becomes corrupted or otherwise comes to lose data.

Abstract: Providing differing degrees of reliability or security for distinct devices within a reliable distributed system. Allowing virtual machines to operate in a reliable distributed system without either exposing the actual available devices to guest operating systems, or requiring the guest OS to maintain, or to maintain information about, reliability or security. Methods are responsive to a hierarchy of possible reliability or security failures, to guard more carefully against simultaneous failures of reliability or breaches of security, without additional work or delay. Functions invoked by the user that relate to reliability and security are hidden, so the user can specify a degree of reliability or security without having to implement those requirements themselves. Providing a uniform set of resources available to all users, separating out those individual resources that are allocated to particular users, and emulating particular devices at the request of those particular users.

Abstract: Techniques that can minimize latency during remote access between nodes in a reliable distributed computing system. A local computing device includes a protocol data unit that interacts with a substantially similar unit at a remote computing device. Each unit executes a finite state machine that is designed to minimize both the amount of wait time between messages and the number of round-trip message exchanges. The state machine resolves each request as quickly as possible to one of three states: SUCCESS, in which the transfer is successful and it is feasible to free the data transfer buffer, COMPLETE, in which the transfer is finished and a next transfer can be initiated, ABORT, in which either the sending or receiving unit declares the transfer to have failed (or aborted by a higher level element).

Abstract: Managing resources in a VM cluster; allocation of resources among competing VMs. Assigning VMs to real devices, responsive to needs for resources: processor usage, disk I/O, network I/O, memory space, disk space. Assigning VMs responsive to needs for cluster activity: network response latency, QoS. Using predictive models of resource usage by VMs. Transferring VMs, improving utilization. Failing-soft onto alternative VM resources. Providing a resource buffer for collective VM resource demand. Transferring VMs to a cloud service that charges for resources.

Abstract: In a reliable distributed computing system, techniques where user actions or events affecting real system devices (RSD's), or causing the system to make changes to metadata, require fewer accesses to RSD's. When a virtual system device (VSD) is mounted, substantially all the metadata, or at least that associated with data recovery, is read into memory. Changes to that metadata can be made substantially without any metadata accesses to RSD's. No metadata accesses to RSD's are made for read operations, as the RSD would remain unchanged. Metadata is flushed from memory upon write operations. Metadata about data locations can be maintained in memory, so as to be available if any RSD becomes corrupted or otherwise comes to lose data.

Abstract: Real system devices (RSD's) emulate virtual system devices (VSD's); system elements can use RSD's and manipulate VSD's, while users can only use VSD's. Identifying when RSD's have erroneous data. Providing a mapping of RSD's to VSD's, to recover a maximum set of VSD's. Providing a mapping of VSD's to RSD's, to recover a specified most valuable VSD. Reconstructing VSD's from RSD's having erroneous data, and coordinating with a user's assistance to couple RSD's in an order allowing data recovery. Metadata is maintained coupling data identifiable on VSD's with corresponding data on RSD's; it is possible to trace connections between RSD's and VSD's even without complete data.