Hyper-V, part of the Windows Server 2008 R2 platform, provides some improvements that were absolutely necessary for Microsoft to even think of competing with VMware's latest offerings. Are they enough? eWEEK Labs' early look at the new Hyper-V shows that Microsoft still has a lot of ground to cover.
Microsoft released Windows Server 2008 R2 with a newly improved version of Hyper-V. Even so, VMware is still miles ahead in terms of the features and innovation that lay the foundation for sustainable virtualization for midsize and large enterprises.
In fact, I think VMware—with its just-released vSphere 4—has raised the bar so high that Microsoft's best hope is to be the low-cost leader. But while cheap, "You get what you pay for" products might work in a consumer category, they won't play too well in IT shops that depend on high-performance data operations to stay in business.
That said, here's what's new and compelling in Hyper-V.
The previous version of Hyper-V had Quick Migration to move virtual machines from one physical host to another. Now, Quick Migration is gone and Live Migration is here.Resource Library:
Click here for a look at improvements in the Hyper-V implementation.
In the weeks ahead, I'll be conducting extensive Live Migration tests on the Labs' Hewlett-Packard and Sun Xeon 5500 ("Nehalem")-based systems. But, for now, let's just say that Quick Migration was so inferior to VMware's VMotion that Microsoft had to shore up this function in Hyper-V.
I suspect that Live Migration has some catching up to do with similar VMware features that have been in field use for several years. When it comes to failover, high availability and load balancing, there is no substitute for production experience. This is one area in which cheap and OK is trumped by market-priced and reliable.
Cluster Shared Volumes are also improved in this version of Hyper-V and play an important role in making VMs highly available. The fact that these clustering enhancements support Live Migration makes them important, but they are no means innovative.
Included among the improvements is a best-practices tool to help ensure proper system configuration. I'm anxious to get started putting a clustered Hyper-V environment together here in the lab. I'll be making extensive use of this tool to see how helpful it is in putting my storage and computing resources into correct alignment.
Microsoft does have a leg up on VMware in at least one area.
Sometime in the next couple of months, Microsoft will release the next version of its System Center Virtual Machine Manager. Microsoft has years of experience in managing large numbers of Windows systems, as well as an almost equal number of years in working with third-party tool makers. Even though most of Microsoft's management experience is with Microsoft-only tools, this could be the edge it needs to win over the virtualization hearts and minds of IT managers, who will soon be measured on how well they manage their virtualized data centers (if they aren't already).
Look for my review of Hyper-V as part of eWEEK Labs' extensive coverage of the Windows Server 2008 R2 platform and Windows 7.
Technical Director Cameron Sturdevant can be reached at csturdevant@eweek.com.
Wednesday, August 12, 2009
Microsoft Azure Pulling Out of Northwest Due to Taxes
Microsoft Azure, its public cloud-based developer platform, will no longer distribute applications to developers from its northwestern hub, as a "change in local tax laws" compels Microsoft to migrate those applications to other geographies prior to the service’s commercial launch in November. Microsoft hopes that it can seize market-share in the cloud-based services space by persuading developers to adopt the platform quickly.
Microsoft's Azure, its public cloud-based developer platform, will soon offer developers one less geographical region from which to run their applications. Azure relies on a worldwide network of distributed data centers to deliver SAAS (software as a service) to users.
"Due to a change in local tax laws, we’ve decided to migrate Windows Azure applications out of our northwest data center prior to our commercial launch this November," announced an Aug. 4 posting on the official Windows Azure blog. "This means that all applications and storage accounts in the 'USA – Northwest' region will need to move to another region in the next few months, or they will be deleted."
The posting added: "Around the time that the 'USA – Northwest' option is removed, we will also provide an automated tool available on the Windows Azure portal to migrate projects." An e-mail will be sent to CTP participants when the tool becomes available on that as-yet-unannounced date.
In addition to "USA – Northwest," Azure offers "USA – Southwest" and "USA – Anywhere" as geographies from which users can run Azure applications. Microsoft plans on adding further geographies at an undetermined time in the future. Resource Library:
Microsoft announced at its Worldwide Partner Conference in July that Azure would be available for free until this year’s Professional Developers Conference in November. After that point, customers will have three different options for paying for the service: a pay-as-you-go model, a subscription format or via volume licensing.
For all three types of service, users will pay 10 cents per gigabyte for incoming data, and 15 cents for outgoing data. The "consumption" model will cost 12 cents per hour for infrastructure usage, and another 15 cents per gigabyte for storage. The business edition of the SQL Azure database will cost $99.99.
Microsoft’s initial price cuts are designed to build market momentum for the platform, which will face competition from similar cloud-based offerings by Amazon.com and Google. Doug Hauger, general manager of Microsoft Azure, told an audience at the Worldwide Partner Conference that Microsoft would offer discounts for partners, as well as allow partners to charge customers for applications and services built using the platform.
Azure will allow developers to "deliver solutions very, very quickly," Hauger added at the time.
Azure and other cloud-based services still face structural issues, such as unexpected downtime, but analysts also feel that their presence could ease many IT professionals’ reservations about running parts of their operations in a public cloud.
Microsoft has no plans to make Azure available to run in an enterprise’s private cloud. Even so, some of Azure’s functionality, including the ability to boot from a VHD (virtual hard disk) has been integrated into Windows Server 2008.
Microsoft's Azure, its public cloud-based developer platform, will soon offer developers one less geographical region from which to run their applications. Azure relies on a worldwide network of distributed data centers to deliver SAAS (software as a service) to users.
"Due to a change in local tax laws, we’ve decided to migrate Windows Azure applications out of our northwest data center prior to our commercial launch this November," announced an Aug. 4 posting on the official Windows Azure blog. "This means that all applications and storage accounts in the 'USA – Northwest' region will need to move to another region in the next few months, or they will be deleted."
The posting added: "Around the time that the 'USA – Northwest' option is removed, we will also provide an automated tool available on the Windows Azure portal to migrate projects." An e-mail will be sent to CTP participants when the tool becomes available on that as-yet-unannounced date.
In addition to "USA – Northwest," Azure offers "USA – Southwest" and "USA – Anywhere" as geographies from which users can run Azure applications. Microsoft plans on adding further geographies at an undetermined time in the future. Resource Library:
Microsoft announced at its Worldwide Partner Conference in July that Azure would be available for free until this year’s Professional Developers Conference in November. After that point, customers will have three different options for paying for the service: a pay-as-you-go model, a subscription format or via volume licensing.
For all three types of service, users will pay 10 cents per gigabyte for incoming data, and 15 cents for outgoing data. The "consumption" model will cost 12 cents per hour for infrastructure usage, and another 15 cents per gigabyte for storage. The business edition of the SQL Azure database will cost $99.99.
Microsoft’s initial price cuts are designed to build market momentum for the platform, which will face competition from similar cloud-based offerings by Amazon.com and Google. Doug Hauger, general manager of Microsoft Azure, told an audience at the Worldwide Partner Conference that Microsoft would offer discounts for partners, as well as allow partners to charge customers for applications and services built using the platform.
Azure will allow developers to "deliver solutions very, very quickly," Hauger added at the time.
Azure and other cloud-based services still face structural issues, such as unexpected downtime, but analysts also feel that their presence could ease many IT professionals’ reservations about running parts of their operations in a public cloud.
Microsoft has no plans to make Azure available to run in an enterprise’s private cloud. Even so, some of Azure’s functionality, including the ability to boot from a VHD (virtual hard disk) has been integrated into Windows Server 2008.
DataCore, Citrix Combine to Distribute Free VSAN Appliance
The Virtual SAN Appliance will enable IT shops that use Citrix Systems virtual server tools to test out DataCore Software's storage software at no cost. The Virtual SAN Appliance evaluation software connects Citrix Essentials with a StorageLink-certified, iSCSI shared-storage SAN that can pool up to 1TB of disk space.
Storage virtualization and disaster recovery provider DataCore Software said Aug. 6 that it has joined with Citrix Systems to distribute a free virtual storage area network appliance for IT managers to test out.
The Virtual SAN Appliance software will enable IT shops that use Citrix's virtual server tools to test out DataCore's storage software at no cost. Resource Library:
Click here to read about DataCore's SAN starter packages.
DataCore said Citrix has "released a Citrix StorageLink adapter to integrate its storage virtualization solutions with Citrix Essentials for XenServer and Microsoft Hyper-V. The Virtual SAN Appliance evaluation software ... [connects] Citrix Essentials with a StorageLink-certified, iSCSI shared-storage SAN that can pool up to 1TB of disk space."
It should be no surprise that none of this will work with VMware's ESX hypervisor, since Citrix is a dyed-in-the-wool competitor of VMware, the world's largest virtualization software company.
The virtual appliance also features thin-provisioning storage for virtual machines, speeds up application performance and creates snapshots for "Instant Volume Cloning and Fast Disk-to-Disk Backups," DataCore said.
Thin provisioning is a method of storage resource management and virtualization that lets IT administrators limit the allocation of physical storage to what applications immediately need.
Thin provisioning also enables the automatic addition of capacity on demand up to preset limits so that IT departments can avoid buying and managing excessive amounts of disk storage.
The DataCore Virtual SAN Appliance and StorageLink Adapter can be downloaded
Storage virtualization and disaster recovery provider DataCore Software said Aug. 6 that it has joined with Citrix Systems to distribute a free virtual storage area network appliance for IT managers to test out.
The Virtual SAN Appliance software will enable IT shops that use Citrix's virtual server tools to test out DataCore's storage software at no cost. Resource Library:
Click here to read about DataCore's SAN starter packages.
DataCore said Citrix has "released a Citrix StorageLink adapter to integrate its storage virtualization solutions with Citrix Essentials for XenServer and Microsoft Hyper-V. The Virtual SAN Appliance evaluation software ... [connects] Citrix Essentials with a StorageLink-certified, iSCSI shared-storage SAN that can pool up to 1TB of disk space."
It should be no surprise that none of this will work with VMware's ESX hypervisor, since Citrix is a dyed-in-the-wool competitor of VMware, the world's largest virtualization software company.
The virtual appliance also features thin-provisioning storage for virtual machines, speeds up application performance and creates snapshots for "Instant Volume Cloning and Fast Disk-to-Disk Backups," DataCore said.
Thin provisioning is a method of storage resource management and virtualization that lets IT administrators limit the allocation of physical storage to what applications immediately need.
Thin provisioning also enables the automatic addition of capacity on demand up to preset limits so that IT departments can avoid buying and managing excessive amounts of disk storage.
The DataCore Virtual SAN Appliance and StorageLink Adapter can be downloaded
Riverbed Offers Virtualized WAN Optimization Management
Riverbed, which has already virtualized its Steelhead Mobile Controller appliance, is now offering a virtualized version of its Central Management Console, enabling MSPs to centrally manage multiple Steelhead appliances from a single physical server. The CMC-VE will let MSPs save on both capital and operational expenses, with further savings offered through flexible licensing.
Weeks after rolling out a virtualized version of its Steelhead Mobile Controller appliance, Riverbed Technology is offering managed service providers an easier way to manage multiple sets of Riverbed’s Steelhead WAN optimization appliances.
Riverbed Aug. 10 unveiled its Central Management Console-Virtualized Edition—or CMC-VE—enabling MSPs to centrally manage and scale multiple Steelhead appliances for their customers from a single physical machine, eliminating the need for additional hardware.
“They can support multiple deployments of Steelheads from a single console,” Karin Gerster, director of product management for verticals for Riverbed, said in an interview.Resource Library:
With IT budgets continuing to tighten, MSPs are looking for ways to control costs and spend more wisely, Gerster said. The CMC-VE will enable MSPs to do just that, she said.
While the number of Steelhead appliances that can be managed from a single physical server varies depending on the MSPs’ environments, one internal test showed that one server was able to run 50 CMC-VE instances, with each instance managing 50 Steelhead appliances. Another internal test had the same server hosting six CMC-VE instances, with each instance managing 500 appliances.
Using the CMC-VE, MSPs are able to save their clients both extra capital and operational expenses, Gerster said. Until now, the capabilities in the CMC were offered only through hardware appliances.
CMC-VE runs on VMware’s ESX virtualization platform, and can run on any server running that software.
In addition, Riverbed is offering flexible licensing, enabling MSPs to buy additional management licenses for Steelhead appliances as they’re needed. In addition, they can essentially create a pool of licenses for Steelhead appliances, and can move existing licenses from customer to customer as needed.
Weeks after rolling out a virtualized version of its Steelhead Mobile Controller appliance, Riverbed Technology is offering managed service providers an easier way to manage multiple sets of Riverbed’s Steelhead WAN optimization appliances.
Riverbed Aug. 10 unveiled its Central Management Console-Virtualized Edition—or CMC-VE—enabling MSPs to centrally manage and scale multiple Steelhead appliances for their customers from a single physical machine, eliminating the need for additional hardware.
“They can support multiple deployments of Steelheads from a single console,” Karin Gerster, director of product management for verticals for Riverbed, said in an interview.Resource Library:
With IT budgets continuing to tighten, MSPs are looking for ways to control costs and spend more wisely, Gerster said. The CMC-VE will enable MSPs to do just that, she said.
While the number of Steelhead appliances that can be managed from a single physical server varies depending on the MSPs’ environments, one internal test showed that one server was able to run 50 CMC-VE instances, with each instance managing 50 Steelhead appliances. Another internal test had the same server hosting six CMC-VE instances, with each instance managing 500 appliances.
Using the CMC-VE, MSPs are able to save their clients both extra capital and operational expenses, Gerster said. Until now, the capabilities in the CMC were offered only through hardware appliances.
CMC-VE runs on VMware’s ESX virtualization platform, and can run on any server running that software.
In addition, Riverbed is offering flexible licensing, enabling MSPs to buy additional management licenses for Steelhead appliances as they’re needed. In addition, they can essentially create a pool of licenses for Steelhead appliances, and can move existing licenses from customer to customer as needed.
VMware and SpringSource: It`s All About the Cloud, Baby
Though Facebook's acquisition of FriendFeed seemed to garner most of the headlines on Aug. 10, the bigger story for the enterprise was easily VMware's announcement of its intent to acquire SpringSource for $362 million to bolster its cloud strategy against Microsoft and others.
Though Facebook's acquisition of FriendFeed seemed to garner most of the headlines on Aug. 10, the bigger story for the enterprise was easily VMware's announcement of its intent to acquire SpringSource for $362 million to bolster its cloud strategy against Microsoft and others.
Make no mistake, that is what's at the core of VMware's move here. With SpringSource under its wing, VMware can become the Java-based equivalent to what is expected to be Microsoft's Azure private cloud play -- which has .NET as its development platform. But rather than .NET, VMware will have the Java-based Spring Framework and its surrounding set of Eclipse-based tools as the development environment for the emerging VMware vCloud private cloud initiative.
Microsoft will likely leverage its Azure, .NET, Hyper-V, System Center, and PowerShell, as well as Windows Server and other software to deliver its private cloud play. In a blog post, Microsoft’s Steven Martin talks a bit about this. More on Microsoft’s plans is expected at the Microsoft Professional Developers Conference (PDC) in November in Los Angeles.
Meanwhile, VMware will likely leverage its vSphere, Spring Framework and tools, SpringSource Hyperic system management software and other secret sauce virtualization technologies to deliver the VMware private cloud solution.Resource Library:
In an interview with eWEEK, Mik Kersten, a Java developer, CEO of Tasktop and creator of the Eclipse Mylyn project, explained his view of the tooling part of the equation:
"In the past year Tasktop has been working closely with SpringSource on the cloud deployment tooling in the SpringSource Tool Suite (STS). That has given me a perspective on how important virtualization is about to become to a large number of Java developers. Consider how the success of the Windows client and servers followed an investment in developer tools, APIs and the programming model for deploying to physical hardware. Virtualized infrastructure is replacing that physical hardware as the new deployment target, and Spring already has the hearts and minds of the majority of Java developers. With the acquisition, Spring developers can expect to start leveraging the profound capabilities of a virtualized infrastructure, while the tool support in STS will make it easy for them to deploy and manage applications in the cloud. This marks a big step forward for enterprise Java."
Indeed, a recent Evans Data survey indicated that a majority of developers chose Java as the top language for building private cloud applications, followed by Microsoft's C#.
Meanwhile, in a separate interview with eWEEK, Rod Johnson, CEO of what will become the SpringSource division of VMware, said, "When you look at the large enterprise market, which is our bread and butter, every CIO [Chief Information Officer] is trying to figure out their cloud strategy and what they can do behind their firewall to build their private cloud. And if they're in the enterprise they’re likely to be using Java. And if they're using Java they're likely to be using Spring. So you take the Java programming model with Spring and you combine it with our management offering (Hyperic) and the VMware virtualization platform. This gives the virtualization layer deep knowledge of what's happening inside the application at runtime."
Moreover, this integration builds naturally as the management software serves to promote a more "autonomic" and instrumented environment, and will enable the virtualization layer "to make decisions in real time" about things such as how many servers to add to support larger workloads, Johnson said. Other such decisions include: Which components need to be co-located on the same virtual server to minimize latency? What layer of the application or infrastructure is the bottleneck? Will increasing compute resources solve the problem, or are there other gating factors?
To the question of whether the VMware marriage with SpringSource is aimed primarily at Microsoft, Johnson told eWEEK:
"There's an industry wide shift toward cloud and the concept of Platform as a Service (PAAS). Microsoft clearly recognizes this. It's no longer about the operating system. As we said when we launched our Build/Run/Manage initiative at SpringSource a few months ago, Microsoft is one of the few companies with a credible, joined up message from developer desktop to data center and cloud. I see the relationship as one of ‘coopetition.’ There are areas in which SpringSource works together with Microsoft right now, and will potentially work together in the future, and we're keen to continue that as a division of VMware. We'll also continue our commitment to Spring.NET, which is growing in popularity."
So not only does Microsoft have reason for concern about this VMware/SpringSource PAAS play from a Java perspective, but perhaps the software giant might also have need for a casual concern from a .NET perspective to the extent that Spring developers begin to show interest in .NET (or .NET developers to show interest in Spring).
But that is further from the focus of this deal.
Moreover, not only does this deal put VMware in more competition with Microsoft, but it puts the virtualization juggernaut in the sights of Amazon, Salesforce.com and Google as well. Microsoft has .NET, Google has Python -- and to some extent Java -- but with this move, VMware is making a power play to claim the title of being the application platform for Java-based cloud development. If nothing else, this VMware deal could serve to stall CIO decisions about which direction to take in the cloud as the various contenders continue to build out their platforms.
Though Facebook's acquisition of FriendFeed seemed to garner most of the headlines on Aug. 10, the bigger story for the enterprise was easily VMware's announcement of its intent to acquire SpringSource for $362 million to bolster its cloud strategy against Microsoft and others.
Make no mistake, that is what's at the core of VMware's move here. With SpringSource under its wing, VMware can become the Java-based equivalent to what is expected to be Microsoft's Azure private cloud play -- which has .NET as its development platform. But rather than .NET, VMware will have the Java-based Spring Framework and its surrounding set of Eclipse-based tools as the development environment for the emerging VMware vCloud private cloud initiative.
Microsoft will likely leverage its Azure, .NET, Hyper-V, System Center, and PowerShell, as well as Windows Server and other software to deliver its private cloud play. In a blog post, Microsoft’s Steven Martin talks a bit about this. More on Microsoft’s plans is expected at the Microsoft Professional Developers Conference (PDC) in November in Los Angeles.
Meanwhile, VMware will likely leverage its vSphere, Spring Framework and tools, SpringSource Hyperic system management software and other secret sauce virtualization technologies to deliver the VMware private cloud solution.Resource Library:
In an interview with eWEEK, Mik Kersten, a Java developer, CEO of Tasktop and creator of the Eclipse Mylyn project, explained his view of the tooling part of the equation:
"In the past year Tasktop has been working closely with SpringSource on the cloud deployment tooling in the SpringSource Tool Suite (STS). That has given me a perspective on how important virtualization is about to become to a large number of Java developers. Consider how the success of the Windows client and servers followed an investment in developer tools, APIs and the programming model for deploying to physical hardware. Virtualized infrastructure is replacing that physical hardware as the new deployment target, and Spring already has the hearts and minds of the majority of Java developers. With the acquisition, Spring developers can expect to start leveraging the profound capabilities of a virtualized infrastructure, while the tool support in STS will make it easy for them to deploy and manage applications in the cloud. This marks a big step forward for enterprise Java."
Indeed, a recent Evans Data survey indicated that a majority of developers chose Java as the top language for building private cloud applications, followed by Microsoft's C#.
Meanwhile, in a separate interview with eWEEK, Rod Johnson, CEO of what will become the SpringSource division of VMware, said, "When you look at the large enterprise market, which is our bread and butter, every CIO [Chief Information Officer] is trying to figure out their cloud strategy and what they can do behind their firewall to build their private cloud. And if they're in the enterprise they’re likely to be using Java. And if they're using Java they're likely to be using Spring. So you take the Java programming model with Spring and you combine it with our management offering (Hyperic) and the VMware virtualization platform. This gives the virtualization layer deep knowledge of what's happening inside the application at runtime."
Moreover, this integration builds naturally as the management software serves to promote a more "autonomic" and instrumented environment, and will enable the virtualization layer "to make decisions in real time" about things such as how many servers to add to support larger workloads, Johnson said. Other such decisions include: Which components need to be co-located on the same virtual server to minimize latency? What layer of the application or infrastructure is the bottleneck? Will increasing compute resources solve the problem, or are there other gating factors?
To the question of whether the VMware marriage with SpringSource is aimed primarily at Microsoft, Johnson told eWEEK:
"There's an industry wide shift toward cloud and the concept of Platform as a Service (PAAS). Microsoft clearly recognizes this. It's no longer about the operating system. As we said when we launched our Build/Run/Manage initiative at SpringSource a few months ago, Microsoft is one of the few companies with a credible, joined up message from developer desktop to data center and cloud. I see the relationship as one of ‘coopetition.’ There are areas in which SpringSource works together with Microsoft right now, and will potentially work together in the future, and we're keen to continue that as a division of VMware. We'll also continue our commitment to Spring.NET, which is growing in popularity."
So not only does Microsoft have reason for concern about this VMware/SpringSource PAAS play from a Java perspective, but perhaps the software giant might also have need for a casual concern from a .NET perspective to the extent that Spring developers begin to show interest in .NET (or .NET developers to show interest in Spring).
But that is further from the focus of this deal.
Moreover, not only does this deal put VMware in more competition with Microsoft, but it puts the virtualization juggernaut in the sights of Amazon, Salesforce.com and Google as well. Microsoft has .NET, Google has Python -- and to some extent Java -- but with this move, VMware is making a power play to claim the title of being the application platform for Java-based cloud development. If nothing else, this VMware deal could serve to stall CIO decisions about which direction to take in the cloud as the various contenders continue to build out their platforms.
Penguin Offers Cloud Computing for HPC
Linux cluster vendor Penguin Computing has created a cloud computing environment aimed at the HPC space. Penguin’s POD service is built on the vendor's Intel-powered Linux clusters, high-speed interconnect technologies like InfiniBand, NetApp SANs, Nvidia graphics chips and Penguin’s Scyld ClusterWare management software, all important technologies for highly parallel, memory-intensive HPC applications. Penguin also is not using virtualization technologies on its server clusters, which officials said will improve server and I/O performance.
Penguin Computing is creating a cloud environment specifically for the high-performance computing space.
Penguin Aug. 11 announced its POD (Penguin on Demand) service that is aimed at researchers, scientists and businesses that need extra compute power for their HPC applications or organizations that want HPC capabilities without having to incur the acquisition and maintenance costs of the cluster hardware.
Charles Wuischpard, president and CEO of the Linux cluster vendor, said Penguin officials saw a need for a cloud computing environment that was made for HPC applications. Popular public clouds, like Amazon’s EC2 (Elastic Compute Cloud) are good for the transaction-intensive, general-purpose applications, but are not made for the highly parallel tasks that are found in HPC environments.
“The workloads are different than what is on Amazon,” Wuischpard said in an interview.Resource Library:
Penguin has built a cloud computing service based in a colocation center in Utah that is made for those HPC workloads. POD is built out of highly optimized Linux clusters that are powered by about 1,200 processing cores of Intel’s Xeon 5400 processors. It also takes advantage of high-speed interconnect technologies, including InfiniBand and Gigabit Ethernet, and a high-speed NAS (network attached storage) from NetApp.
POD also comes with Penguin workstations powered by Nvidia’s Tesla graphics processors. In addition, Penguin is not virtualizing its servers, as is normally done at places like Amazon’s EC2. Instead, Penguin officials decided to offer customers all the resources of a server to increase performance and give businesses access to all the I/O, which is important when running the highly parallel, memory-intensive HPC workloads, Wuischpard said.
The result is that HPC workloads run as much as 30 times faster on the Penguin cloud computing service than on EC2. For example, one workload that takes 18 hours on EC2 can be completed in 30 minutes on Penguin’s POD.
“Amazon does an excellent job of catering to server-based workloads,” said Josh Bernstein, HPC architect and a principal designer of the POD technology for Penguin. However, its EC2 is not built for HPC tasks.
The entire environment is managed by Penguin’s Scyld ClusterWare software, and the vendor offers a variety of support services, including application set up, creating an HPC environment and ongoing maintenance. Penguin also is working with software makers who will offer their HPC software on POD.
POD is the latest step for Penguin officials in their efforts to focus the company on HPC. Since taking over the reins of the company two years ago, Wuischpard has shed a lot of Penguin’s ancillary Linux computing business to focus on HPC, a move he said has resulted in strong growth during that time.
Penguin’s POD is a further example of how Wuischpard and other officials are looking to expand the company’s reach.
“The way it changes our business is that it takes us out of the pure hardware game,” he said.
Penguin Computing is creating a cloud environment specifically for the high-performance computing space.
Penguin Aug. 11 announced its POD (Penguin on Demand) service that is aimed at researchers, scientists and businesses that need extra compute power for their HPC applications or organizations that want HPC capabilities without having to incur the acquisition and maintenance costs of the cluster hardware.
Charles Wuischpard, president and CEO of the Linux cluster vendor, said Penguin officials saw a need for a cloud computing environment that was made for HPC applications. Popular public clouds, like Amazon’s EC2 (Elastic Compute Cloud) are good for the transaction-intensive, general-purpose applications, but are not made for the highly parallel tasks that are found in HPC environments.
“The workloads are different than what is on Amazon,” Wuischpard said in an interview.Resource Library:
Penguin has built a cloud computing service based in a colocation center in Utah that is made for those HPC workloads. POD is built out of highly optimized Linux clusters that are powered by about 1,200 processing cores of Intel’s Xeon 5400 processors. It also takes advantage of high-speed interconnect technologies, including InfiniBand and Gigabit Ethernet, and a high-speed NAS (network attached storage) from NetApp.
POD also comes with Penguin workstations powered by Nvidia’s Tesla graphics processors. In addition, Penguin is not virtualizing its servers, as is normally done at places like Amazon’s EC2. Instead, Penguin officials decided to offer customers all the resources of a server to increase performance and give businesses access to all the I/O, which is important when running the highly parallel, memory-intensive HPC workloads, Wuischpard said.
The result is that HPC workloads run as much as 30 times faster on the Penguin cloud computing service than on EC2. For example, one workload that takes 18 hours on EC2 can be completed in 30 minutes on Penguin’s POD.
“Amazon does an excellent job of catering to server-based workloads,” said Josh Bernstein, HPC architect and a principal designer of the POD technology for Penguin. However, its EC2 is not built for HPC tasks.
The entire environment is managed by Penguin’s Scyld ClusterWare software, and the vendor offers a variety of support services, including application set up, creating an HPC environment and ongoing maintenance. Penguin also is working with software makers who will offer their HPC software on POD.
POD is the latest step for Penguin officials in their efforts to focus the company on HPC. Since taking over the reins of the company two years ago, Wuischpard has shed a lot of Penguin’s ancillary Linux computing business to focus on HPC, a move he said has resulted in strong growth during that time.
Penguin’s POD is a further example of how Wuischpard and other officials are looking to expand the company’s reach.
“The way it changes our business is that it takes us out of the pure hardware game,” he said.
Numonyx CTO Lays Out Road Map for Phase-Change Memory
Phase-change memory, which stands to replace DRAM and other components in digital devices during the next five years, offers a great deal of promise for the entire IT industry. It is blazingly fast, reliable and versatile. However, it is expensive, and in the current economic climate it has been a tough sell as a replacement technology.
SANTA CLARA, Calif.—Numonyx CTO Ed Doller laid out the differences in latencies between spinning disk hard drives, solid-state NAND flash drives and phase-change memory in terms that a lay person can readily understand.
"There are some very big differences, and processes in the future are only going to be demanding more speed," Doller told a near-capacity audience of several hundred people Aug. 11 in a keynote at the fourth annual Flash Memory Summit here at the Convention Center.
"I live in the world of milliseconds, nanoseconds and microseconds, but to normalize things, I'm going to make this more familiar to a wider audience," Doller said.
"Here it is: If I assume that random latency in a PCM device is one day—let's say it takes me one day to go to my PCM, get the information randomly and bring it back. In a solid-state drive, it [that same operation] would take 17 days to get that information back. How does that compare to a hard disk drive? Nine years!"
Doller then went a step further. "Let's say I'm going to put my PCM interface right on my processor, and I'm going to use PCM as storage-class memory—which means I'm going to execute code directly out of the PCM. Well, then it [the latency] goes from one day to 30 minutes," Doller said.Resource Library:
Therein lies the value of phase-change memory: speed that conventional data storage drives have never seen before—and never will.
A phase-change memory chip—it is also known as PRAM—is nonvolatile memory that works well for both executing code and storing large amounts of data, giving it a superset of the capabilities of both flash memory and dynamic random access memory. This means it can execute code with performance, store larger amounts of memory and also sustain millions of read/write cycles.
Intel debuted the first PCM chips at its developers' conference in San Francisco in September 2006.
The wafer shown to eWEEK that day represented Intel and Italy's ST Microelectronics' first grasp of the new type of nonvolatile memory chip. The two companies later merged to create Numonyx.
A great deal of development has been completed at Numonyx since then; Doller said that the adoption by mainstream IT companies has been slow but that it would take use by only a couple of big names—Apple and Microsoft would be two of them—for PCM to take off into the market stratosphere.
"PCM is on the verge, and we think it's inevitable that it will replace a lot of what is in devices today," Doller said. "But to start, it would take a company that is good at producing both hardware and software to make best use of it. It's bound to happen."
Adoption Curve Slow but Expected to Speed Up
Doller thinks PCM will need three to five more years to attain widespread adoption. Right now, the biggest drawback in PCM is the price, which is about 10 times higher than DRAM at this point. The pricing, however, will come down over time and as fabricating processes become improved.
PCM chips use the same material, chalcogenide, that is used inside to store data in rewritable optical disks. But instead of using a laser to change the properties of the material and thus create the zeros and ones that make up data, the chips use electricity that flows through a resistor. The resistor heats up and does the job of the laser, changing the materials' properties to represent a zero or a one.
The effort is the "culmination of [work by] some of the smartest materials guys on the planet," Doller said. "Over the years, this has an opportunity to be a very large memory technology."
Most industry people and analysts strongly believe PCM has the potential to replace both NAND—flash memory designed primarily for data storage—and NOR flash memory, designed for executing code with one type of chip, streamlining manufacturing processes.
PCM: 'Nirvana' for an Engineer
"PCM is like a superset of NOR or NAND flash," Doller said. "It's almost nirvana for an engineer. It reads fast, writes fast—it does everything faster."
Doller said the kind of speed PCM brings is going to be needed three to five years from now "just to index the Internet."
"We're bringing up a new generation that is used to instant-on," Doller said. "They expect to get things instantly.
"For example, my daughter was texting people on her iPhone instead of doing her homework recently. I asked her to stop texting her friends and go into her room, get on the PC and do her homework. She said, 'Dad, I am doing my homework.' Turns out she was doing it on her iPhone. She didn't want to use the PC because it took too long to spin up."
SANTA CLARA, Calif.—Numonyx CTO Ed Doller laid out the differences in latencies between spinning disk hard drives, solid-state NAND flash drives and phase-change memory in terms that a lay person can readily understand.
"There are some very big differences, and processes in the future are only going to be demanding more speed," Doller told a near-capacity audience of several hundred people Aug. 11 in a keynote at the fourth annual Flash Memory Summit here at the Convention Center.
"I live in the world of milliseconds, nanoseconds and microseconds, but to normalize things, I'm going to make this more familiar to a wider audience," Doller said.
"Here it is: If I assume that random latency in a PCM device is one day—let's say it takes me one day to go to my PCM, get the information randomly and bring it back. In a solid-state drive, it [that same operation] would take 17 days to get that information back. How does that compare to a hard disk drive? Nine years!"
Doller then went a step further. "Let's say I'm going to put my PCM interface right on my processor, and I'm going to use PCM as storage-class memory—which means I'm going to execute code directly out of the PCM. Well, then it [the latency] goes from one day to 30 minutes," Doller said.Resource Library:
Therein lies the value of phase-change memory: speed that conventional data storage drives have never seen before—and never will.
A phase-change memory chip—it is also known as PRAM—is nonvolatile memory that works well for both executing code and storing large amounts of data, giving it a superset of the capabilities of both flash memory and dynamic random access memory. This means it can execute code with performance, store larger amounts of memory and also sustain millions of read/write cycles.
Intel debuted the first PCM chips at its developers' conference in San Francisco in September 2006.
The wafer shown to eWEEK that day represented Intel and Italy's ST Microelectronics' first grasp of the new type of nonvolatile memory chip. The two companies later merged to create Numonyx.
A great deal of development has been completed at Numonyx since then; Doller said that the adoption by mainstream IT companies has been slow but that it would take use by only a couple of big names—Apple and Microsoft would be two of them—for PCM to take off into the market stratosphere.
"PCM is on the verge, and we think it's inevitable that it will replace a lot of what is in devices today," Doller said. "But to start, it would take a company that is good at producing both hardware and software to make best use of it. It's bound to happen."
Adoption Curve Slow but Expected to Speed Up
Doller thinks PCM will need three to five more years to attain widespread adoption. Right now, the biggest drawback in PCM is the price, which is about 10 times higher than DRAM at this point. The pricing, however, will come down over time and as fabricating processes become improved.
PCM chips use the same material, chalcogenide, that is used inside to store data in rewritable optical disks. But instead of using a laser to change the properties of the material and thus create the zeros and ones that make up data, the chips use electricity that flows through a resistor. The resistor heats up and does the job of the laser, changing the materials' properties to represent a zero or a one.
The effort is the "culmination of [work by] some of the smartest materials guys on the planet," Doller said. "Over the years, this has an opportunity to be a very large memory technology."
Most industry people and analysts strongly believe PCM has the potential to replace both NAND—flash memory designed primarily for data storage—and NOR flash memory, designed for executing code with one type of chip, streamlining manufacturing processes.
PCM: 'Nirvana' for an Engineer
"PCM is like a superset of NOR or NAND flash," Doller said. "It's almost nirvana for an engineer. It reads fast, writes fast—it does everything faster."
Doller said the kind of speed PCM brings is going to be needed three to five years from now "just to index the Internet."
"We're bringing up a new generation that is used to instant-on," Doller said. "They expect to get things instantly.
"For example, my daughter was texting people on her iPhone instead of doing her homework recently. I asked her to stop texting her friends and go into her room, get on the PC and do her homework. She said, 'Dad, I am doing my homework.' Turns out she was doing it on her iPhone. She didn't want to use the PC because it took too long to spin up."
Subscribe to:
Posts (Atom)