The Experts below are selected from a list of 582 Experts worldwide ranked by ideXlab platform
Sanidhya Kashyap - One of the best experts on this subject based on the ideXlab platform.
-
Opportunistic Spinlocks: Achieving Virtual Machine Scalability in the Clouds
Operating Systems Review, 2016Co-Authors: Sanidhya KashyapAbstract:With increasing demand for big-data processing and faster in-memory databases, cloud providers are moving towards large Virtualized Instances besides focusing on the horizontal scalability. However, our experiments reveal that such Instances in popular cloud services (e.g., 32 vCPUs with 208 GB supported by Google Compute Engine) do not achieve the desired scalability with increasing core count even with a simple, embarrassingly parallel job (e.g., Linux kernel compile). On a serious note, the internal synchronization scheme (e.g., paraVirtualized ticket spinlock) of the Virtualized Instance on a machine with higher core count (e.g., 80-core) dramatically degrades its overall performance. Our finding is different from the previously well-known scalability problem (i.e., lock contention problem) and occurs because of the sophisticated optimization techniques implemented in the hypervisor---what we call sleepy spinlock anomaly. To solve this problem, we design and implement OTICKET, a variant of paraVirtualized ticket spinlock that effectively scales the Virtualized Instances in both undersubscribed and oversubscribed environments.
-
Scalability in the Clouds! A Myth or Reality?
2015Co-Authors: Sanidhya Kashyap, Changwoo Min, Taesoo KimAbstract:With increasing demand of big-data processing and faster in-memory databases, cloud providers are gearing towards large Virtualized Instances rather than horizontal scalability. However, our experiments reveal that such Instances in popular cloud services (e.g., 32 vCPUs with 208 GB sup-ported by Google Compute Engine) do not achieve the de-sired scalability with increasing core count even with a sim-ple, embarrassingly parallel job (e.g., kernel compile). On a serious note, the internal synchronization scheme (e.g., par-aVirtualized ticket spinlock) of the Virtualized Instance on a machine with higher core count (e.g., 80-core) dramatically degrades its overall performance. Our finding is different from a previously well-known scalability problem (lock con-tention problem), and occurs because of the sophisticated optimization techniques implemented in the hypervisor, what we call—sleepy spinlock anomaly. To solve this problem, we design and implement oticket, a variant of paraVirtualized ticket spinlock that effectively scales the Virtualized Instances in both undersubscribed and oversubscribed environments. 1
-
APSys - Scalability in the Clouds!: A Myth or Reality?
Proceedings of the 6th Asia-Pacific Workshop on Systems - APSys '15, 2015Co-Authors: Sanidhya KashyapAbstract:With increasing demand of big-data processing and faster in-memory databases, cloud providers are gearing towards large Virtualized Instances rather than horizontal scalability. However, our experiments reveal that such Instances in popular cloud services (e.g., 32 vCPUs with 208 GB supported by Google Compute Engine) do not achieve the desired scalability with increasing core count even with a simple, embarrassingly parallel job (e.g., kernel compile). On a serious note, the internal synchronization scheme (e.g., paraVirtualized ticket spinlock) of the Virtualized Instance on a machine with higher core count (e.g., 80-core) dramatically degrades its overall performance. Our finding is different from a previously well-known scalability problem (lock contention problem), and occurs because of the sophisticated optimization techniques implemented in the hypervisor, what we call---sleepy spinlock anomaly. To solve this problem, we design and implement oticket, a variant of paraVirtualized ticket spinlock that effectively scales the Virtualized Instances in both undersubscribed and oversubscribed environments.
Frank H. P. Fitzek - One of the best experts on this subject based on the ideXlab platform.
-
5G World Forum - Comparing and Evaluating Application-specific Boot Times of Virtualized Instances
2019 IEEE 2nd 5G World Forum (5GWF), 2019Co-Authors: Robert-steve Schmoll, Hani Salah, Tobias Fischer, Frank H. P. FitzekAbstract:With the increasing demand for digital services, requirements for their underlying technologies continue to steadily increase. When originally introduced, virtualization significantly changed the landscape of server provisioning and continues to be the technology of choice for cloud infrastructure. Flexibility has become essential, especially with the explosion of the number of connected devices. One key performance indicator of cloud platforms in regards to flexibility is the time it takes a Virtualized Instance to be up and running on the hardware. While there are studies measuring that time, none of them include the availability of the service inside the Instance. Therefore, we extend the definition of a reachable Instance to include the startup time of the actual service, which is closer to real-world examples. In this work we compare for the first time traditional virtual machines (VMs), container virtualization and unikernels in this new key performance indicator to further understand each technology's applicability in actual scenarios. We show that the type of virtualization has an impact on the overhead of service startup time compared to the boot time of the Instance. Unikernels achieve the lowest values in both boot time and startup time. The best result we observed for a webserver response to a HTTP request is 40 ms, which is achieved by using MirageOS, and is about a tenth of the time a Docker container needs for the same task.
Taesoo Kim - One of the best experts on this subject based on the ideXlab platform.
-
Scalability in the Clouds! A Myth or Reality?
2015Co-Authors: Sanidhya Kashyap, Changwoo Min, Taesoo KimAbstract:With increasing demand of big-data processing and faster in-memory databases, cloud providers are gearing towards large Virtualized Instances rather than horizontal scalability. However, our experiments reveal that such Instances in popular cloud services (e.g., 32 vCPUs with 208 GB sup-ported by Google Compute Engine) do not achieve the de-sired scalability with increasing core count even with a sim-ple, embarrassingly parallel job (e.g., kernel compile). On a serious note, the internal synchronization scheme (e.g., par-aVirtualized ticket spinlock) of the Virtualized Instance on a machine with higher core count (e.g., 80-core) dramatically degrades its overall performance. Our finding is different from a previously well-known scalability problem (lock con-tention problem), and occurs because of the sophisticated optimization techniques implemented in the hypervisor, what we call—sleepy spinlock anomaly. To solve this problem, we design and implement oticket, a variant of paraVirtualized ticket spinlock that effectively scales the Virtualized Instances in both undersubscribed and oversubscribed environments. 1
Robert-steve Schmoll - One of the best experts on this subject based on the ideXlab platform.
-
5G World Forum - Comparing and Evaluating Application-specific Boot Times of Virtualized Instances
2019 IEEE 2nd 5G World Forum (5GWF), 2019Co-Authors: Robert-steve Schmoll, Hani Salah, Tobias Fischer, Frank H. P. FitzekAbstract:With the increasing demand for digital services, requirements for their underlying technologies continue to steadily increase. When originally introduced, virtualization significantly changed the landscape of server provisioning and continues to be the technology of choice for cloud infrastructure. Flexibility has become essential, especially with the explosion of the number of connected devices. One key performance indicator of cloud platforms in regards to flexibility is the time it takes a Virtualized Instance to be up and running on the hardware. While there are studies measuring that time, none of them include the availability of the service inside the Instance. Therefore, we extend the definition of a reachable Instance to include the startup time of the actual service, which is closer to real-world examples. In this work we compare for the first time traditional virtual machines (VMs), container virtualization and unikernels in this new key performance indicator to further understand each technology's applicability in actual scenarios. We show that the type of virtualization has an impact on the overhead of service startup time compared to the boot time of the Instance. Unikernels achieve the lowest values in both boot time and startup time. The best result we observed for a webserver response to a HTTP request is 40 ms, which is achieved by using MirageOS, and is about a tenth of the time a Docker container needs for the same task.
Hani Salah - One of the best experts on this subject based on the ideXlab platform.
-
5G World Forum - Comparing and Evaluating Application-specific Boot Times of Virtualized Instances
2019 IEEE 2nd 5G World Forum (5GWF), 2019Co-Authors: Robert-steve Schmoll, Hani Salah, Tobias Fischer, Frank H. P. FitzekAbstract:With the increasing demand for digital services, requirements for their underlying technologies continue to steadily increase. When originally introduced, virtualization significantly changed the landscape of server provisioning and continues to be the technology of choice for cloud infrastructure. Flexibility has become essential, especially with the explosion of the number of connected devices. One key performance indicator of cloud platforms in regards to flexibility is the time it takes a Virtualized Instance to be up and running on the hardware. While there are studies measuring that time, none of them include the availability of the service inside the Instance. Therefore, we extend the definition of a reachable Instance to include the startup time of the actual service, which is closer to real-world examples. In this work we compare for the first time traditional virtual machines (VMs), container virtualization and unikernels in this new key performance indicator to further understand each technology's applicability in actual scenarios. We show that the type of virtualization has an impact on the overhead of service startup time compared to the boot time of the Instance. Unikernels achieve the lowest values in both boot time and startup time. The best result we observed for a webserver response to a HTTP request is 40 ms, which is achieved by using MirageOS, and is about a tenth of the time a Docker container needs for the same task.
