Supporting metered clients with manycore through time-limited partitioning

[0001] An embodiment of the present invention relates generally to manycore processing environments, and, more specifically, to partitioning resources of a manycore platform for simultaneous use by multiple clients, or additional capacity to a single client.

[0002] Various mechanisms exist for supporting multiple clients on a single computing device. Some legacy methods on main frames included timesharing of clients, where only one process was actually running at a time, but each client's process received a share of available processing time. Other methods for multiple clients to operate on a single computing device include using virtual machines with a virtual machine monitor to control devices and active processes.

[0003] An existing model for sharing a computing device among many users is known as the Internet Café model. In this business method, a number of computing devices are co-located in a café or other business establishment. A user pays for use, often limited by time, and logs on to the computing device. When the user is finished, the computing device may be rebooted in preparation for another user to log on and use the resource. Rebooting, however, is inefficient as it requires significant downtime of the resource, over time. As computing resources on a single platform increase, for instance, with multi-core platforms and faster processors, single user platforms will tend to be underutilized in this type of environment.

[0004] The existing Internet Café model entails having a single machine/user. Single user platforms incur manageability costs and difficulty in scaling, for instance, to provide a pedestal personal computer (PC) in front of each user. The cost of the equipment is also another issue. Security of the hardware may be a problem as well, since users are typically co-located with the processor and peripherals used.

[0013] An embodiment of the present invention is a system and method relating to taking advantage of the multi-core technology, also known as manycore, for a platform having typically more than eight processing cores, in order to provide metered client usage for multiple clients. In the context of the description below, a core is to he understood as a processing unit in a processor on a single socket. For instance, a processor within a socket may have eight cores, or compute nodes within the processor. This is to be distinguished from multi-processor systems that have a single processor in a single socket. A platform may be both multi-processor and multi-core. For instance a muiti-processor platform may have four multi-core processors, each multi-core processor in a single socket. In a four socket platform, with each socket having a four-core processor, the effective number of processing units on the platform will be 16. Other business models for simultaneously enabling multiple clients secure access to the platform are described in alternative embodiments.

[0014] In at least one embodiment, the present invention is intended to use out-of-band communication with an administrator or administrative console to partition and assign cores and resources on the platform to multiple users, or clients. For instance, resources that may be assigned to different sessions or clients include: processing cores, memory and I/O devices. In an embodiment, different clients/sessions may be allocated a varying number of resources.

[0015] Reference in the specification to “one embodiment” or “an embodiment” of the present invention means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. Thus, the appearances of the phrase “in one embodiment” appearing in various places throughout the specification are not necessarily all referring to the same embodiment.

[0016] For purposes of explanation, specific configurations and details are set forth in order to provide a thorough understanding of the present invention. However, it will be apparent to one of ordinary skill in the art that embodiments of the present invention may be practiced without the specific details presented herein. Furthermore, well-known features may be omitted or simplified in order not to obscure the present invention. Various examples may be given throughout this description. These are merely descriptions of specific embodiments of the invention. The scope of the invention is not limited to the examples given.

[0017] A metered computing environment in emerging markets, for instance, those that have embraced the Internet Café model, may have centrally-located service support of clients with multi-core capabilities, such as available from Intel Corporation. Several platforms with manycore technology are either planned or will be deployed in the future to better take advantage of embodiments of the present invention. Specifically, embodiments of the invention having many core-based CPU sockets, as available from Intel Corp., may host a series of clients in an Internet Café (Icafé) and optional usage time metering. Another time metering application is available using Microsoft® FlexGo software. More information about this type of metering may be found on the public Internet at URL www*Microsoft*com/whdc/payasyougo/. Note that periods have been replaced with asterisks in URLs in this document to prevent inadvertent hyperlinks.

[0018] Embodiments of the invention allocate system sources on demand, to different client sessions. In existing systems, IBM System i™ Capacity on Demand uses a method by which inactive processing power may be activated, on demand, by a system administrator. Specifically, a platform is shipped with some compute power in a deactivated state. When the platform owner decides that additional capacity is required, the owner pays the vendor for some incremental increase in capacity, up to and including all that is available on the platform. The vendor sends the administrator/owner an authorization code and the administrator enters the code to activate the additional capacity. A reboot may typically be required to activate the additional capacity. However, this method does not allocate previously activated individual cores and network resources to different clients for metered processing, on the fly. Further, embodiments of the present invention may allocate and reallocate cores and other resources, dynamically. More information about IBM System i™ Capacity on Demand may be found on the public Internet at URL www-03*ibm*com/systems/i/hardware/cod/index.html.

[0019] Embodiments of the invention allow for a single platform to support 32+ users with the same memory and processor isolation as a separate machine; in other words, there is no need for a third party virtual machine monitor. Intel® processors to be deployed with 32 cores are representative of the emergent architecture to support embodiments of the invention, with one of the manycores reserved per user. It will be understood that processors having any number of cores in excess of one may be employed in embodiments of the invention. The ability to employ replicated I/O in future input/output controller hubs (ICH's), and the partitioning capability of a platform resource layer, or embedded partition architecture, make embodiments of the present invention viable.

[0020] An embodiment of the invention may utilize a manycore processor with a hardware partition manager to dedicate specific memory, network I/O, and processor resources to a given Internet Café user. Each user merely has access to a simple terminal with a screen and keyboard/mouse input. The central location of the storage and processor resources allows for physical security. A blade server with the manycore processors may be kept in a locked cabinet, in the Icafé, thus obviating concerns about the user stealing computing system components.

[0021] An advantage of embodiments of the invention arises from the physical security of isolated processors and memory and NIC/client session. This contrasts with existing server consolidation using an operating system (OS) process/client (brittle kernel-mode sharing) or virtual machine monitor (VMM) architecture.

[0022] FIG. 1 is a block diagram illustrating an exemplary topology for the multi-user manycore platform with an out-of-band microcontroller (OOB μcontroller), according to embodiments of the invention. A platform 200 comprises a processor 201. The processor 201 may be connected to random access memory 205 via a memory controller hub 203. Processor 201 may be any type of processor capable of executing software, such as a microprocessor, digital signal processor, microcontroller, or the like. Though FIG. 1 shows only one such processor 201, there may be one or more processors in the platform 200 and one or more of the processors will include multiple cores, or the like.

[0023] The processor 201 may be further connected to I/O devices via an input/output controller hub (ICH) 207. The ICH may be coupled to various devices, such as a super I/O controller (SIO), keyboard controller (KBC), or trusted platform module (TPM) via a low pin count (LPC) bus 202. The SIO, for instance, may have access to floppy drives or industry standard architecture (ISA) devices. In an embodiment, the ICH is coupled to non-volatile memory via a serial peripheral interface (SPI) bus 204. The non-volatile memory may be FLASH memory or static random access memory (SRAM), or the like. An out-of-band (OOB) μcontroller 210n may be present on the platform 200. The OOB μcontroller 210n may connect to the ICH via a bus 212, typically a peripheral component interconnect (PCI) or PCI express bus. The OOB μcontroller may also be coupled with the non-volatile memory store (NV store) 217 via the SPI bus 204. The NV store 217 may be FLASH memory or static RAM (SRAM), or the like. In many existing systems, the NV store is FLASH memory.

[0024] The OOB μcontroller 210n may be likened to a “miniature” processor. Like a full capability processor, the OOB controller has a processor unit 211 which may be operatively coupled to a cache memory 215, as well as RAM and ROM memory 213. The OOB μcontroller may have a built-in network interface 227 and independent connection to a power supply 225 to enable out-of-band communication even when the in-band processor 201 is not active.

[0025] In embodiments, the processor has a basic input output system (BIOS) 219 in the NV store 217. In other embodiments, the processor boots from a remote device (not shown) and the boot vector (pointer) resides in the BIOS portion 219 of the NV store 217. The OOB controller 210n may have access to all of the contents of the NV store 217, including the BIOS portion 219 and a protected portion 221 of the non-volatile memory. In some embodiments, the protected portion 221 of memory may be secured with Intel® Active Management Technology (iAMT). In an embodiment, the portion 221 of the NV store is protected from access by the firmware based on chipset selections in a base address register (BAR). More information about iAMT may be found on the public Internet at URL www*intel*com/technology/manage/iamt/.

[0026] Since the BIOS portion 219 of non-volatile memory may be modified by the OS or applications running within the OS, it may be vulnerable to malicious tampering. The protected area of memory 221, available only to the OOB μcontroller, may be used to store critical boot vector, or other information, without risk of tampering. The only way to access the OOB μcontroller side of the NV store 217 is through verification via a proxy through the OOB μcontroller, i.e., signature authentication or the like. Embodiments of the present invention may utilize a hardware protected region 221 of the non-volatile memory 217 and make the protected region inaccessible to the OS.

[0027] In some embodiments, a CMOS chip 223 accommodates systems with legacy firmware. In these legacy systems, boot settings may be stored on the CMOS in proprietary formats, rather than in a NV store like FLASH memory. A snapshot of CMOS settings may be stored in a known location. Some embodiments may have FLASH memory 217 and CMOS 23.

[0028] In an embodiment, implementation of “mailboxes” to pass messages and data between the in-band (host processor communication) and out-of-band processor may be according to techniques discussed in U.S. patent application Ser. No. 10/964,355 (Attorney Docket: P19896), entitled “BUS COMMUNICATION EMULATION” to Rothman et al. and filed on Oct. 12, 2004.

[0029] The OOB μcontroller 210n may be operated to store a “message” containing a directive in a memory shared by the OOB μcontroller 210n and a processor of the computer system such as the processor 201 of the host computer 200. In the illustrated embodiment, the host computer 200 includes a shared memory 252 which is accessible by both the processor 201 and the OOB μcontroller 210n. The shared memory 252 may reside in a reserved area of RAM 252a, or be located in a separate non-volatile memory store 252b, or the like. The shared memory may be operated as a mailbox for these messages. Thus, in one aspect, the OOB μcontroller 210n may store a message in the shared memory 252 or retrieve a message from the shared memory 252, independently of the status of the host computer 200 including the status of the processor 201, the operating system and the programs. Thus, in the illustrated embodiment, the OOB μcontroller 210n may store or retrieve messages in the shared memory 252 whether the Processor 201 is being initialized or is turned off, or whether the operating system is booting, running, crashed or otherwise.

[0030] To facilitate such independent operation, in this example, the controller 210n, the shared memory 252, the local bus 212 and other components as appropriate may be powered independently of the main components of the host computer 200 including the processor 201 and the host memory 205. The shared memory 252 may be non-volatile (NV) memory such as FLASH memory or static random access memory (SRAM). In embodiments described in more detail below, the OOB μcontroller 210n operates independently of the operating system or system start-up program, such that the OOB μcontroller 210n may have its own dedicated control circuitry, firmware, operating system, etc. to control the operations of the OOB μcontroller 210n independently of the status of the remainder of the host computer 200. It is appreciated that the degree of operational independence, if any, of the controller 210n and other components may vary, depending upon the particular application.

[0031] In embodiments of the present invention, the OOB μcontroller 210n communicates, out-of-band, with an administrator or administrative console program to reallocate resources of the manycore processor based on client demand.

[0032] FIG. 2 is a block diagram illustrating the manycore architecture of a platform, according to embodiments of the invention, and how the cores are allocated among multiple users. While only two user/client sessions are illustrated, it will be understood that there may be as many client sessions as there are available cores or resources on the platform. Platform 100 is illustrated with 32 cores 110, but it will be understood that embodiments of the invention may be implemented on platforms with any number of cores (more than one). The cores 110 are coupled to a memory controller hub (MCH) 120 which enables control and communication with system memory 130. A “cross-section” of system memory 130 is shown to illustrate that multiple client sessions may have access to only a partition of memory. For instance, client session 1 has access only to system memory partition 131. Client session 2 has access only to system memory partition 132. The memory is fully isolated and secure between and among client sessions, as more fully described below.

[0033] As illustrated with the solid arrowed line that starts at isolated memory 131 and travels to client session 1 (151), four cores 111 are partitioned for use by session 1. As illustrated with the broken arrowed line that starts at isolated memory 132 and travels to client session 2 (152), four cores 113 are partitioned for use by session 2. 24 cores (115a-c) are unused in this example, and thus, available for additional client sessions. Sessions 1 and 2 are shown as being connected to network 150 via network adapters 145 coupled to multiple input/output hubs (IOHs) of input/output controller hubs (ICHs) 140. Each client session may have a dedicated network controller hub 145. In some embodiments, the clients may connect to the platform wirelessly.

[0034] FIG. 3A is a block diagram illustrating the communication and control by the out-of-band processor, which may also be called a “manageability engine” (ME). In this illustration, the manycore processor 300 is shown with cores (processing units), as well as L3 cache. Cache adjacency to assigned cores may be important to policy models which may automatically allocate resources, based on demand and current, availability. In an embodiment, an administrator 310, or in some cases, an automated administrative console 311, receive requests from users/clients who desire a session on the platform. In some cases, the request is a face to face request to be administered by the administrator. In another embodiment, a client may log onto the administrative console and make an electronic request, and even pay for metered time via a credit or debit card at the console.

[0035] The administrator (and console) is communicatively coupled to an out of band processor (ME) 330 via an IT network 320. The ME may have a number of internal RMII (Reduced Media-Independent Interface) buses and connections to various devices on the platform, for instance a double data rate 2 (DDR2) bus to a DDR2 memory device. The ME is coupled to the manycore processor 300 and controls the partitioning of resources into various secure client, environments. It will be understood that the ME may be coupled to the manycore directly or via an internal proprietary bus. As illustrated, there is a memory and I/O layer 301 around the cores and 1.3 caches of the manycore. This layer represents connections to external memory and the I/O may be PCI express (PCIe) buses. The memory and I/O controllers are not shown in this figure. The PCIe buses may connect directly from the chip. A direct memory interconnect (DMI) bus may be used to connect the chip to a peripheral controller hub (PCH). A PCH may be used instead of an ICH. Other architectures may be used, based on chip voltages and other platform parameters.

[0036] Effectively, the resources need to be virtualized, or partitioned, such that a collection of cores are to be dedicated to a single client or user. Referring again to FIG. 2, the cores are partitioned as a four-core session 111, a 12-core session or spare, 115c, a six-core session or spare 115a, a six-core session or spare 115b, and a four-core session 113. The hardware is programmed, including but not limited to, the links/fabric/network between the processors, in order to segregate different sets of cores, memory and I/O. In this illustration, the platform is segregated into the equivalent of five individual processing entities. Once configured, each processing entity, or session, will have dedicated memory in system memory 130, a dedicated network, adapter 145, etc. In existing systems this type of segregation would be defined before boot, and remain static. In some embodiments, the platform allows only one user and session, for instance, a desktop model, and unused cores are unavailable to the user until a further metered payment is made. However, in embodiments of the present invention, cores may be dynamically allocated and active cores and inactive securely partitioned from each other. Once the user is authorized to access additional cores, they are made available via a repartitioning.

[0037] Referring again to FIG. 3A, the ME 330 enables the manycore platform to be managed externally to change the configuration of cores and resources within the platform, on demand. In one embodiment, the platform adheres to a policy based partitioning, i.e., the policy may be based on a pay-as-you-go model, where one or more users pay for a finite amount of compute power (number of cores and memory) and a finite amount of compute time. If a user's workload increases or decrease, the user may request a readjustment of resources, and a like readjustment of the fee for those metered resources. The ME. 330 communicates with the chipset of the platform to initiate the changes in resource partitioning.

[0038] Each core 350 in the manycore chipset 300 has a router, or route table 351, that that identifies which other cores it is connected to at any given time. Each core also has source address decoder (SAD) 352 which tells the core which portions of memory it may use. Only those portions of memory programmed into the SAD are accessible to the core. Each core may also have a mailbox register 353 to hold information received from the ME and to be processed. In some embodiments, the core is an ×86 processing unit. The ME 330 programs the core's on-processor route tables to define the current connections. Super privileged firmware working in concert with the ME 330 programs the on-processor routers, configuration registers and memory controller 120 that only the firmware can access. The terra configuration register may be used in this description to represent interfaces/elements that software accesses to manage the SAD/Router/mailbox elements; e.g., any SMM or ME-visible hardware that is changeable. An operating system executing on a core has no access to these elements of the chipset. The co-processor rooter is configured to define which core is connected to which other core, which core is connected to which memory partition, and which cores are connected to which I/O devices. In manycore platforms, the cores are connected with control and data lines as compared to other types of networks, but the connections are built-in to the chipset. The cores and memory controllers are connected with registers that only the platform firmware can access. The article entitled “Research Challenges For On-chip Interconnection,” by John D. Owens, et al, (IEEE Computer Society. 0272-1732/07 2007), provides some additional information about the interconnections between and among the cores. This article may also be found on the public Internet at URL doi*ieeecomputersociety*org/10.1109/MM.2007.91.

[0039] FIG. 3B is a block diagram further illustrating the control and communication among cores, according to an embodiment of the invention FIG. 3B illustrates selected aspects of a manycore computing system having access control logic, according to an embodiment of the invention. System 300 includes a number of cores 312 which are interconnected by routing logic 314. Cores 312 may be general purpose processor cores, graphics cores, and the like. It is to be appreciated that system 300 may include nearly any number of cores (e.g., 2, 4, 8, 16, 32, 64, 128, etc.).

[0040] Routing logic 314 may include the source address decoders (SADs) 352 and/or route tables 351 that are used to interconnect cores 312. Various aspects of routing logic 314 may be configurable. For example, how physical addresses are decoded may be configurable and/or the values in the route tables may be configurable.

[0041] In some embodiments, at least some of the instances of routing logic 314 include access control logic 316 and control status register (CSR) 318. Access control logic 316 may determine whether an agent is authorized to change the values stored in CSR 318. In some embodiments, access control logic 316 requires that an agent seeking to access CSR 318 provides a public key matching a private key that was previously provisioned within system 300. In alternative embodiments, a different mechanism may be used to authenticate an agent. In some embodiments, access control logic 316 uses a cryptographic authentication protocol to authenticate an agent. In some embodiments, access control logic 316 is implemented, at least in part, in hardware. In alternative embodiments, access control logic 316 may be implemented, at least in part, in platform microcode.

[0042] In an embodiment, the platform further comprises a peer to peer (pTp) interconnect technology, also sometimes called common system interface (CSI) architecture. A CSI, or pTp, system processor may include an integrated memory controller to reduce the memory access latency. Server systems may now be built using two to hundreds of such processors. In such systems, only one processor executes BIOS code to initialize memory and other portions of the system. However, as the number of processors and memory controllers has increased, the mechanisms to initialize the system and boot an operating system commonly take extended periods. As a result, initialization may be parallelized between the multiple processors.

[0043] A pTp is a reprogrammable interconnect, where the reprogramming is by a distinguished platform agent to provide, gather and separate different pools of computing resources on a platform.

[0044] In some pTp based multi-processor systems, all of the processors execute the BIOS code in parallel and perform a Built-in Self Test (BIST), memory initialization, and the like to speed up the system boot time. These activities are coordinated by a Boot Strap Processor (BSP) to collect the data from each of the other processors, which may be referred to as Application Processors (AP), and boot the operating system. Parallel initialization increases the computation requirements to initialize memory and pTp links. As a result, processor cache is configured as a temporary data store, which may be referred to as Cache-As-RAM (CAR), to the processor socket in order to assist in the memory, BIST execution, and pTp initialization.

[0045] However, the temporary data store in one processor cannot be accessed by other processors In particular, the BSP cannot directly access the AP temporary stores to create global Source Address Decoder (SAD) entries to represent the whole system. Previous efforts to allow the BSP to create the global SAD include using a Configuration Space Register (“CSR”), a chipset or non-core register accessible by both BSP and AP processors, as a mailbox to exchange data.

[0046] In an embodiment, a platform may have multiple processors configured to communicate using pTp technology, and each processor in a single socket has many cores configured, as discussed above.

[0047] FIG. 4 is a logical block diagram of a system 400, according to an exemplary embodiment. The system 400 includes four processing units 402, 406, 410 and 414. Processor 406 is the bootstrap processor (BSP), The processors 402, 406, 410, 414 each include a local memory subsystem 404, 408, 412, and 416, respectively. The system 400 also includes two input/output hubs IOH 0 420 and IOH 1 428. The input/output hubs 420, 428 provide connectivity to input/output devices, such as input/output controller hub (ICH) 422 and PCI Express 424, 426. Processor-to-processor and processor-to-input/output hub 420, 428 communication may be performed using pTp packets. Each pTp component contains a Routing Table Array (RTA) and a SAD. The RTA provides the pTp packet routing information to other sockets. The SAD provides mechanisms to represent routing of the resources such as memory, input/output, and the like. Each processor 402, 406, 410, 414 also contains a Target Address Decoder (TAD). The TAD provides mechanisms to map system addresses to processor 402, 406, 410, 414 memory 404, 408, 412, 416 addresses, respectively.

[0048] In a link-based multi-processor system, such as the system 400, after the system 400 is reset, the Basic Input/Output System (BIOS) starts to execute from each processor 402, 406, 410, 414 in a parallel manner. Each processor 402, 406, 410, 414 performs its own power on self test (POST) and initializes the components in its respective domain, such as local links, local memory, and attached local devices. Initialization of each processor 402, 406, 410, 414 also detects error conditions. The BIOS logs this processor 402, 406, 410, 414, which may also be referred to as a socket, information into local processor storage, such as cache-as-RAM (CAR) or local memory 404, 408, 412, 416 depending on the boot stage and the particular embodiment.

[0049] However, during system 400 boot, one of the processors 402, 406, 410, 414 is selected to control the whole system 400 boot. This processor is referred to as a Bootstrap Processor (BSP). In this example processor 406 is the BSP. The other processors 402, 406, 410, 414 are called Application Processors (AP).

[0050] Major tasks performed by the BSP include discovering the system 400 topology and setup of RTAs such that all components may communicate with each and synchronizes the socket-based information recorded by other processors 402, 406, 410, 414. The BSP also performs the remaining system level initializations, such as setup of the system address map, initialization of input/output devices, and the like. The BSP may also be tasked with loading an operating system on the system 400.

[0051] The synchronization step is important. The synchronization steps performs many functions including getting information including memory organization, DIMM size, memory size, and error information, to construct a global system 400 address map and it determines how to distribute the memory across the system 400. A common way to synchronize information before the system 400 memory addresses are setup is using a scratch register as a mailbox. The mailbox may include two scratch registers—one scratch register as command register through which the BSP and APs may communicate command information, and the other to exchange data. Other communication mechanisms may be used.

[0052] Embodiments of the invention allow dynamic reallocation of cores on a single socket processor to the equivalent of virtual machines, or user sessions/environments. The core's control logic and registers define the interaction of the core with other cores, and with memory. By reprogramming the cores' configuration, many permutations of environments (sessions) may be generated to provide a variable multi-user environment, or provide a single environment with varying processing capacity.

[0053] In an embodiment, a message is received by the ME to configure a six core session with 1 GB of memory and two network controllers. The ME may then generate a system management interrupt (SMI) within the system management mode (SMM) to reprogram the routing controllers. In other embodiments, the route tables are written by microcode, or other constructs. The SMM code has access to the routing controller registers, but the operating system(s) (OS) do not. Concurrent partitions, or sessions, may run simultaneously. To each partition, it seems that the partition is a unique machine (e.g., a virtual machine, or session). The OS running on the partition is isolated from other partitions and has no direct access to the chipset or router controllers to modify the platform configuration. The OS also has no access to memory or resources that are dedicated to other partitions, or sessions.

[0054] In an embodiment, cache memory, as shown as L3 blocks in the manycore 300, may be allocated to a partition by being allocated to a specific core.

[0055] Once the console or administrator is directed to authorize resources for a session, it must be determined whether sufficient cores are available to allocate; this calculation may be culled via active management technology (AMT) reporting the availability of “unassigned” cores which the ME can track via a bit-map of assigned versus unassigned processor elements. In an embodiment, the console sends a web services simple object access protocol (SOAP) message to the ME to learn how many cores are unallocated. The ME may send an SMI to a chipset register with a command to request information about unallocated cores and resources. The ME effectively sends an in-band message to the BIOS or firmware to determine the information. If the ME needs to communicate with an OS, an interrupt may be generated. If the ME needs to communicate with the firmware, an SMI may be generated once availability is determined. The ME reserves the appropriate resources by generating an SMI event into a register to write into the route tables of the cores. The SMM reads the information in the message register and performs the appropriate writes to route tables.

[0056] In some implementations, there may be physical limits regarding which cores may be combined in a session, or which caches may be connected to specific cores. The ME may have a predetermined platform policy that dictates these limitations and drives the generation and allocation of sessions and partitions. In some cases, sufficient adjacent cores may not be available, or the available cores may not have cache connectivity. In this case, in an embodiment, the ME may use platform policy to determine whether a requested configuration is possible given currently allocated resources, and whether the resources may be reallocated to provide for a session that would not otherwise be available.

[0057] In an embodiment, when the ME determines that a reallocation is required to generate a newly requested session, a command to stop all processing on allocated cores may be sent from the ME to the chipset. The elements in question are to be added or removed and all address decoders and routers are to be reprogrammed—even for cores that are currently allocated to a session and processing applications. This action is analogous to reliability, availability and serviceability (RAS) activities on processors for hot plugging redundant units. However, in this case, the reallocation is dynamically performed for load balancing per customer request and not reliability.

[0058] The cores' application data and state may be stored into a globally-available memory space, such as System Management RAM (SMRAM) or sent to the ME and stored in temporary memory location of the ME. For more transactional aspects (i.e., to survive a power-cycle) the state may be stored by SMM to a system FLASH memory or the ME to its sequestered/protected FLASH prior to the reallocation operation, along with setting a “reallocation in process” flag around the movement of state.

[0059] FIG. 5 is a flow diagram illustrating the platform initialization and session allocation process, according to an embodiment of the invention. The platform is restarted in block 501 The platform is initialized in block 503. This initialization may vary depending on the architecture of the platform. For instance, the platform may be implemented with a peer to peer interconnect (pTp) to support a layered protocol scheme. Once initialized, if there is only one core available, as determined in block 505, then the platform is a legacy platform to be configured as a local host, in block 507. Legacy mode is commenced in block 509.

[0060] If the platform has multiple cores, then it is determined whether metering support is implemented with many networks in block 511. If so, then it is determined whether hardware partitioning, for instance with an embedded partition is implemented, in block 513. If hardware partitioning is available, then the hardware, memory and network partitions are created in block 515. The route tables and SADs are initialized, perhaps to a default partitioning. If a last known good configuration has been saved, the initialization may be to this last known good configuration. During this initialization, the firmware dictates which partition configuration should be used. If no hardware partitioning is implemented, then it is determined whether virtualization is supported in block 519. While virtualization partitioning may be possible, it is not further discussed below. For both virtualization and hardware partitioning, a remote console is then notified of the availability of the resources in the platform, in block 517. In an embodiment, this notification is effected by the BIOS generating a message to the ME, and the ME generating a web services message to send to the console (310/311). In all multi-core scenarios, metered client processing mode is entered in block 521.

[0061] When a new client enters the icafé, or other environment, the client requests a specific amount of computational capability to an administrator or an automated system. In an automated model, the client may pay for the metered usage via credit or debit card right at the console. It will be understood that various methods may be used to enter the information, such as manual keyboard input, swiping a credit, card or wirelessly reading a smartcard, etc. In the case of a human interface, the administrator enters the information regarding requested resources into a console. For purpose of simplification, both the automated method and human interface point of origin will be referred to as the “administrative console.” The administrative console sends a request to the platform via the ME to determine whether cores are available to assign to the client, in block 527. If cores are available, the cores and associated memory, cache and I/O devices are assigned, as described above, and the route tables, and SADs are appropriately assigned, in block 529. The client may then begin using a dumb terminal (I/O device) assigned to the client session.

[0062] If the requested number of cores, or other assets, are not immediately available, the client request may time-out, i.e., be put on a queue of clients requesting allocations that are not yet available. It is determined whether other client sessions have timed out (i.e., used up their allotted metered resources/time), in block 531. If not, processing continues at block 523 and will loop until resources are available for the requested client session. Alternatively, platform policy is compared to allocated cores and a rebalancing of platform assets may be initiated to enable the client session to be created.

[0063] When there is one or more client session that has exhausted its metered resources, idle cores, memory, cache and I/O devices may be reallocated to the new client session, in block 533.

[0064] If there is no pending new client request, the system monitors to identify idle sessions, in block 525. If there are idle sessions (i.e., a session that is not using the allocated resources) and if a timed-out request is pending, as determined in block 535, then the new session may be allocated in block 537. The cores, and other resources, are reallocated in block 539 When there are no timed-out requests for a session pending, processing continues at block 523 to wait for new requests. Because some client sessions may require fewer resources during human interaction, it may be possible to reallocate cores between one or more client, sessions on the fly, and still be able to deliver the requested processing capacity. This may be analogized to overbooking the resources. However, in contrast to overbooking a commercial airline carrier where excess passengers are bumped to a next flight, overbooked client sessions basically have their resources swapped back and forth during idle times. Caution must be taken before implementing a platform policy allowing overbooking, however, to ensure that sufficient idle time will be available to effectively swap cores and other resource without noticeably impacting on user performance. In other cases, idle sessions refer to sessions that, are no longer active, and the resources may be reallocated at will.

[0065] Load balancing of resources and swapping idle resources between and among client sessions requires a secure memory to temporarily store processing data for a client session, as discussed above. This memory store may be coupled to the ME and operating systems on the manycore processor will have no access to it.

[0066] In another embodiment, a user purchases a platform having n cores. The user pays for m number of cores, where l≧m≦n, and m cores are to be activated. When the system is configured, m cores and associated resources are activated. Thus, the route fables of the cores are configured to identify the active resources in the platform. The user may only use the activated resources. If the user determines that additional resources are needed, additional payments can be made to the vendor and the vendor may “upgrade” the platform by reprogramming the core's route tables, SADs, etc. The platform may be configured such that only the vendor, from a remote terminal, has authorization to reprogram the cores. The platform may be configured so that a key must be entered by a local administrator to activate the additional cores and resources. Regardless, the SMI to force the reprogramming of the route tables, etc., is to be executed only when the vendor has authorized the upgrade. It will be understood that various techniques may be used to safeguard this command. Similar incremental configuration cost models may have been used in existing systems for processor and memory upgrades, but none have been used to activate individual cores on a single socket processor.

[0067] In an embodiment, the platform has an embedded partition architecture to create secure hardware partitions of processors and other resources. This architecture is also known as a platform resource layer (PRL) architecture. In a platform resource layer (PRL) architecture, or embedded partition architecture, various components of the platform are enhanced to enable partitioning of processor, memory and other resources.

[0068] Referring now to FIG. 6, there is shown an exemplary block diagram of a platform 600 having a PRL architecture, according to embodiments of the invention. This is an architecture instance that is usable in an Internet Café wherein each individual processor instance may be assigned to a given Cafe patron. To better illustrate partitioning, components that are available to the main partition 610 are drawn with solid blocks. Components available to the embedded, or system partition 620, are drawn with bold, solid blocks. Components available to both partitions are drawn with a block alternating with dots and dashes.

[0069] In this exemplary embodiment, a platform has four multi-core processors in Sockets 0-3 (631-634). While this example shows only four processor sockets, it will be apparent to one of ordinary skill in the art that various configurations of processors and cores may be used to practice embodiments of the invention. For instance. Socket 0 (631) may have four processing cores 635a-d. In essence, in this example, the illustrated embodiment has 16 effective processors on the platform (e.g., four sockets with four cores in each socket). In this example. Sockets 0-2 (631-633) are available only to the main partition 610. Socket 3 (634) is available to both the main partition 610 and to the embedded partition 620. Within Socket 3 (634), core 0 is available only to the main partition 610, and cores 1-3 are available only to the embedded partition 620. The embedded partition 620 may have a reallocation agent 621 as part of the firmware to communicate with ME, via mailbox messages, to effect the reprogramming of the cores. SMM is a global capability and it can reach out and access any socket. The reallocation agent may contain authorization code to ensure that the vendor has authorized allocation of additional cores and resources.

[0070] In this embodiment, the platform has a memory controller hub (MCH) 601 (also known as north bridge) coupled to memory 602. Memory 602 may have two partitions MEM1 (603) and MEM2 (605). Memory partition MEM1 (603) is available only to the embedded partition and memory partition MEM2 (605) is available only to the main partition. The chipset containing the MCH 601 is configured to partition the memory using hardware constructs, in contrast to a VMM solution which uses software constructs It will be understood that memory 602 may be a hard disk, a floppy disk, random access memory (RAM), read only memory (ROM), flash memory, or any other type of medium readable by processor. Memory 602 may store instructions for performing the execution of embodiments of the present invention. While only two partitions are shown in this example, it will be understood that there may be more than one guest OS, each running in its own partition. It will also be understood that each single processor socket may have any number of cores.

[0071] The MCH 601 may communicate with an I/O controller hub (ICH) 607, also known as South bridge, via a peripheral component interconnect (PCI) bus. The ICH 607 may be coupled to one or more components such as PCI hard drives, legacy components such as IDE, USB, LAN and Audio, and a Super I/O (SIO) controller via a low pin count (LPC) bus (not shown). In this example, the ICH 607 is shown coupled to a hard disk drive 609 and to a network interface controller (NIC) 611.

[0072] The MCH 601 is configured to control accesses to memory and the ICH 607 is configured to control I/O accesses. In an embedded partition architecture, the chipset is configured by the firmware, upon boot, to partition the various resources on the platform. In some cases, there may be only one partition and the platform acts like a legacy platform in most respects. In the example shown, there are two partitions, a main partition 610 and an embedded partition 620. Each partition designated is given a unique partition identifier (ID).

[0073] With an embedded partition configuration, when a device sends an alert, the chipset may properly route the alert to the appropriate partition, as this information is encoded at boot time. In a VMM enabled system, the hardware passes the device alerts to the VMM (virtualized devices) and the software routes the information appropriately to the various virtual machines. An embedded partition may act as hardware assisted virtualization.

[0074] In an embodiment, a reallocation agent is embodied within a VMM which controls all guest virtual machines (VMs) and guest operating systems (OS's) running on the platform. In another embodiment, the reallocation is embodied in a privileged partition, process or hypervisor that controls I/O requests for individual OS's.

[0075] The techniques described herein are not limited to any particular hardware or software configuration; they may find applicability in any computing, consumer electronics, or multi-core processing environment. The techniques may be implemented in hardware, software, or a combination of the two.

[0076] For simulations, program code may represent hardware using a hardware description language or another functional description language which essentially provides a model of how designed hardware is expected to perform. Program code may be assembly or machine language, or data that may be compiled and/or interpreted. Furthermore, it is common in the art to speak of software, in one form or another as taking an action or causing a result. Such expressions are merely a shorthand way of stating execution of program code by a processing system which causes a processor to perform an action or produce a result.

[0077] Each program may be implemented in a high level procedural or object-oriented programming language to communicate with a processing system. However, programs may be implemented in assembly or machine language, if desired. In any case, the language may be compiled or interpreted.

[0078] Program instructions may be used to cause a general-purpose or special-purpose processing system that is programmed with the instructions to perform the operations described herein. Alternatively, the operations may be performed by specific hardware components that contain hardwired logic for performing the operations, or by any combination of programmed computer components and custom hardware components. The methods described herein may be provided as a computer program product that may include a machine accessible medium having stored thereon instructions that may be used to program a processing system or other electronic device to perform the methods.

[0079] Program code, or instructions, may be stored in, for example, volatile and/or non-volatile memory, such as storage devices and/or an associated machine readable or machine accessible medium including solid-state memory, hard-drives, floppy-disks, optical storage, tapes, flash memory, memory sticks, digital video disks, digital versatile discs (DVDs), etc., as well as more exotic mediums such as machine-accessible biological state preserving storage. A machine readable medium may include any mechanism for storing, transmitting, or receiving information in a form readable by a machine, and the medium may include a tangible medium through which electrical, optical, acoustical or other form of propagated signals or carrier wave encoding the program code may pass, such as antennas, optical fibers, communications interfaces, etc. Program code may be transmitted in the form of packets, serial data, parallel data, propagated signals, etc., and may be used in a compressed or encrypted format.

[0080] Program code may be implemented in programs executing on programmable machines such as mobile or stationary computers, personal digital assistants, set top boxes, cellular telephones and pagers, consumer electronics devices (including DVD players, personal video recorders, personal video players, satellite receivers, stereo receivers, cable TV receivers), and other electronic devices, each including a processor, volatile and/or non-volatile memory readable by the processor, at least one input device and/or one or more output devices. Program code may be applied to the data entered using the input device to perform the described embodiments and to generate output information. The output information may be applied to one or more output devices. One of ordinary skill in the art may appreciate that embodiments of the disclosed subject matter can be practiced with various computer system configurations, including multiprocessor or multiple-core processor systems, minicomputers, mainframe computers, as well as pervasive or miniature computers or processors that may be embedded into virtually any device. Embodiments of the disclosed subject matter can also be practiced in distributed computing environments where tasks or portions thereof may be performed by remote processing devices that are linked through a communications network.

[0081] Although operations may be described as a sequential process, some of the operations may in fact be performed in parallel, concurrently, and/or in a distributed environment, and with program code stored locally and/or remotely for access by single or multi-processor machines. In addition, in some embodiments the order of operations may be rearranged without departing from the spirit of the disclosed subject matter. Program code may be used by or in conjunction with embedded controllers.

[0082] While this invention has been described with reference to illustrative embodiments, this description is not intended to be construed in a limiting sense. Various modifications of the illustrative embodiments, as well as other embodiments of the invention, which are apparent to persons skilled in the art to which the invention pertains are deemed to lie within the spirit and scope of the invention.