1295014 ,14565twf.doc/g 九、發明說明： 【發明所屬之技術領域】 本發明是有關於一種對送往資源元件進行排程請求的 I置與方法’且特別是有關於一種根據請求紀錄（hist⑴ 對送往資源元件的請求進行排程的裝置與方法。 【先前技術】 把σ午夕糸統建立在一顆晶片上並以系統中的一此獨立 的初始器（例如··微處理器、信號處理機等）存取二個動 態，機存取記憶體（DRAM)子系統是一種常見的事。對 於每一個初始器（initiator)而言，系統可能需要不同的傳 輸，務品質。其次，記憶體排序模型的方式對於起始器是 =常重要的。理想上，初始器會希望存取一個盡可能有固 疋順序的記憶體模型。同時，這些對DRAM的請求被傳送 到子系統中的順序，將會對dram的表現度有復 :的影響。然而，因服務品質（qualities 〇f service)資‘ 是基於DRAM運作效率的理由再次請求排序時，須 記憶體模型進行妥協，所需要是一個統一的 非耘機制，來表示成一個固定的排序記憶體模型， 能保持不同的服務品質’並使D罐盡可 列（Hi每始器的請求串流可以被描述成串 白因一虫）如果一個dram排程器並未再重新排序來 接W列之多個請求時’内部串列的請求的排序就會唯 列之&13_鳩請求的排序就僅是連續的各請求串 4而已。現有的系統可能在dram效率排程 1295014 J4565twf.doc/g (若有的話）之外的其他場所來排序請求，且/或重新排序 處理串列中的請求，例如：請求可以由一個初始器經由一 個標準的電腦匯流排到DRAM控制器。在存取電腦匯流排 時建立一個請求排序（在串列與串列間），而且在存取中 不允許DRAM控制器改變排序順序。其它的例子如：在美 國專利公告1^〇.6,578，117中被提出的〇以八]^排程，其用於 建立一個高效率的dram及服務品質的保證，在'U7 一 案中，一個_列服務品質排程器保持並使用串列狀態去記 住串列排程紀錄，並有助於決定串列之下一步。舉例說明： 排程器會試圖發出存取相同DRAM分頁的排程請求，這 樣可以增加命中DRAM分頁的機會。 σ 在更進一步的應用上，佇列可被用來儲存請求，且每 個^丁列被分配的優先權都不相同’以致於在高優先件列中 的請求會先被執行，直到所有較高優先請求都被^行完 畢，低優先佇列的請求才被執行。此基於優先權的排程= 廣泛的被使用，因為比較重要的請求優先處理，其表 會比一般排程方法的表現度還要好。 又 然而，此基於優先權的排程器仍舊有一些缺點，i 一個缺點是低優先請求可能會阻礙高優先請求。當=、 始器隨機的發出請求時，就可能會出現將所有的：佟 求執行完畢而後開始執行低優先請求的狀況。然 新的高優先請求抵達而低優先請求還在執行時，此 權請求必須等到低優先請求執行完畢才能開始 =可知，高優綠求會被延遲，且整體缸作效率= 1295014 、14565twf.doc/g 【發明内容】 從某一方面來看，本發明提供了 一種對送往資源元件 的請求進行排程的裝置與方法。此裝置包含一個高優先請 求佇列（queue)，用以儲存多個送往該資源元件（resource device)的高優先請求；一個低優先請求佇列用於儲存一 個送往資源元件的低優先請求，其中，每一高優先請求的 優先順序，會高於低優先請求的優先順序；一個紀錄計數 験器（history counter)用於儲存一個資訊，此資訊相關於至 少一個介於兩相鄰的高優先請求間之請求間隔·，以及一個 排程模組，根據前述的資訊以排序高優先請求及低優先請 求0 就另一方面來說，本發明提供一種對送往資源元件的 請求進行排程的方法，包含··接收多個高優先請求與一個 低優先請求，其中，每一該些高優先請求的優先順序，會 高於低優先請求的優先順序；儲存一資訊，而其資訊相關 • 於至少一個介於兩相鄰的高優先請求間之請求間隔；以及 根據此資訊以排序高優先請求及低優先請求。 依照本發明的較佳實施例所述，其中當該資訊指出有 足夠的時間用於低優先請求之第一部份運作時，則低優先 請求會被分割成第一部分及至少另一部分，且低優先 中的第一部分會被准予請求資源元件。 一 依照本發明的較佳實施例所述，其資訊包含多個請求 間隔的平均值，其中平均值是將每一請求間隔加以權重後 再相加而得。 1295014 14565twf.doc/g 因為本發明考慮了發生在高優先佇列中的請求間隔並 依此資訊排序低優先請求，因此將使得低優先請求可以被 准予進入高優先請求之請求間隔中，且高優先請求被低優 先请求阻斷的機會也將因為低優先請求的特殊排序方式而 降低。 為讓本發明之上述和其他目的、特徵和優點能更明顯 易懂，下文特舉較佳實施例，並配合所附圖式，作詳細說 I 明如下。 【實施方式】 請參照圖1，其繪示依照本發明較佳實施例的一種裝 置方塊圖，在本實施例中，上述的裝置10包含了一個高優 先佇列12、一個低優先佇列14、一個紀錄計數器16及一 個排程模組18。代理者100〜106分享資源元件11〇中的資 源藉由對裝置10發佈請求，比方說一個動態隨機存取記憶 體（dram)。此裝置10接收到多個請求並且排序接收到 > 的請求，去讓這些代理者100-106存取資源元件。 在本實施例中，代理者100及102所發佈的請求，被 蒐集在上述的高優先佇列12中，而代理者1〇4和1〇6所發 出的請求則被t集於低優先佇列14中。在高優先仔列12 t的請求擁有比低優絲列14中的請求相對來說較高的 優先權。進一步的說，只有在高優先佇列12被清空之後， 低優先佇列14中的請求才能得到執行的許可。也就是說， 代理者104和106在存取資源元件11〇時只有較低的優先 權。對於資源元件no來說，必須要妥善規劃由代理者1〇4 1295014 、14565twf.doc/g 與l〇6發出的請求所使用的時間長度。不 能會導致高優先代理者度可 者的資料存取操作效率低落。 /頓或低優先代理 為了決定—個適當的週期長度，紀錄計數 相關於介於兩相鄰高優先請求（亦即，- =12中的請求）之間的請求間隔的資訊。特二知 ，紀=賴ϋ 16自先_高優先請求 "數有 ί少時脈經過，當-個新的高優先請求抵達時，'= 态16儲存凊求間隔的計數結 ^ 然而，紀錄_ 16除Α二啟二—次的計數。 的資訊。舉個例子，前述二特其他 這些請求間隔的平均值。^ =隔或是 :本=權Γ:求，:相加得到請“ ^ 時間。x 包3於則述貧訊中的各種資料被稱為可利用 18根據館存在記錄計數器财的資訊來排 ==先㉔求及低優先請求。首先，排程模組18同意在高 ίίΓ2之切請求進行運作，而在高優先㈣u被 /月工之ί排私杈、组18會開始排序在低優先仔列14中的 請求。 特定一點來說’當儲存在記錄計數器16中的資訊指出 有足夠的時_來進行低錢請求的操作時 ，則此低優先 請求就會被准傾作。細，如果此資訊指出沒有足夠的 l295〇l4 、14565twf.doc/g 時間用來進行-個完整的低優先請求的操作時， 組18會判斷低優先請求的第一部份是否可被准予操^如 果有足夠的時間用於進行低優先請求之第一部分操作日士 ，優先請求會被分誠第—部份和至少另—部份，且:優 先请求之第一部份可以被准予請求資源元件。 請參照圖2 ’其繪示本發明較佳實施例的一種排程方 f流程圖’此排财法可輯用㈣—個㈣元件排程靖 ^。首先’接收包含高優綠求（和/或）低優先請求在内 irf請求（步驟湖）；再來，儲存前述之資訊，而此 貝_關於在兩相鄰高優先請求之間的一個請求間隔 驟幻〇2)，且根據此資訊排序接收的請求（步驟s2 根據本發_較佳實補所述，為了更詳加描述排程 ^ ’睛參相3 ’其描述在圖2步驟S2Q4中更詳細的流 粒圖。 ，本實施财，先檢翅衫個高優絲求的高優先 :列二否被㉔（步驟测）。#高縣剌還沒被清空 :，兩優骑求會被以適當_序准予存㈣源猶（步 担S3〇2)。然而’如果高優先仔列被清空，則會取得上述 &到的資訊（步驟S3〇4)且低優先請求排序因此開始運作。 在取得上述提到的資訊後，此資訊會被檢查而判斷包 =该貧訊内的可利用時間是否有足夠的長度以供進行低 t先請求的操作（步驟S遍）。如果可利用時間夠長足以 進仃-個完整的低優先請求的操作，則低優先請求會被准 1295014 14565twf.doc/g 予進行其對資源元件所請求的操作（步驟S308)。 然而，當可利用時間不足以去進行一個完整的低優先 請求的運作時，就考慮分割低優先請求為幾個部分。舉例 來說，如圖所示之步驟S310，可以先檢查低優先請求以決 疋可以在别述的可利用時間内完成的一個第一部份。如果 可利用日守間不能支持低優先請求的任一部分的操作時，流 程就會回到步驟S300。然而，如果可利用時間足以進行低 優先睛求之第一部分的操作時，此低優先請求會被分割成 兩部分，其中一個部分是前述的第一部份（步驟S312)。 在分割低優先請求之後，低優先請求之第一部份可被准予 2存取資源元件，而低優先請求的另一部份仍停留在低優 4 丁列中’等待下一次准予存取之運作。 件66^此，在本發明中，低優先代理者之請求存取資源元 的』間的適當的長度係藉由最後和即將到來的 理求間隔的評估值決定。本發明最佳化低優先代 能性?；子取的效率，而且最小化高優先請求之停頓的可 限定太f㈣已哺佳實施纖露如上，然其並非用以 和範圍^ i任何熟習此技藝者’在不脫離本發明之精神 範圍=可作些許之更動與潤飾，因此本發明之保護 【圖式簡單說明】 々千 圖1 圖2 ^為本發明較佳#_的—難置方塊圖。 、，曰不為本發明較佳實施例的一種排程方法流程 11 1295014 14565twf.doc/g 圖。 其所 不 圖3繪示為本發明較佳實施例之一種流程圖 係為圖2步驟S204中詳細的流程。 【主要元件符號說明】 10 ·二種對送往資源元件的請求進行排程的 12 ·南優先符列 14 :低優先彳宁列 16 :紀錄計數器 18 :排程模組 100〜106 :代理者 110 :資源元件 S200 :接收高位元請求及低位元請求 S202:判斷-資訊，其資訊 間之請求間隔 7相砷回優先睛求 S204 ^據該資訊排序接收到的請求 S300 :高優先仵列是否被清空 S302 :准予高優先請求 工 S304 :取得該資訊 之第用時間是否夠長足以進行低優先請求 S308 :同意低優先請求 ^行低優先請求 關：判斷可利用時間是 之第一部分的操作 长疋以 =·: $先請求被分割成兩部分 S3M·同意低優先請求之第一部分 121295014, 14565twf.doc/g IX. Description of the Invention: [Technical Field] The present invention relates to an I-position and method for scheduling requests to resource elements, and in particular to a record according to a request ( Hist(1) Apparatus and method for scheduling requests sent to resource elements. [Prior Art] A sigma-day system is built on a single chip and is a separate initializer in the system (for example, a microprocessor) , signal processor, etc.) access to two dynamic, machine access memory (DRAM) subsystem is a common thing. For each initiator, the system may require different transmission, quality. The way the memory is sorted is always important for the initiator. Ideally, the initiator will want to access a memory model that is as solid as possible. At the same time, these requests for DRAM are passed to the subsystem. The order in the order will have a complex effect on the performance of the dram. However, because of the quality of service (qualities 〇f service), it is requested again based on the efficiency of DRAM operation. To compromise the memory model, a unified non-耘 mechanism is needed to represent a fixed sort memory model, which can maintain different service quality' and make D can be listed (Hi every request) Streaming can be described as a string of whites. If a dram scheduler is not reordered to pick up multiple requests for the W column, the ordering of the internal serial requests will be only listed &13_鸠The ordering of the requests is only continuous for each request string 4. Existing systems may sort requests and/or reorder processing at locations other than the dram efficiency schedule 1295014 J4565twf.doc/g (if any). The request in the serial, for example, the request can be queued to the DRAM controller by an initial device via a standard computer. A request ordering (between the serial and the serial) is established when accessing the computer bus, and is stored. The DRAM controller is not allowed to change the sort order. Other examples are: Arranged in US Patent Publication 1 〇.6, 578, 117, which is used to create a highly efficient dram and service. Quality Ensure that in the 'U7 case, a _ column service quality scheduler maintains and uses the serial state to remember the serial schedule record and helps determine the next step in the sequence. Example: Scheduler will Attempting to issue a schedule request to access the same DRAM page, which increases the chance of hitting DRAM pages. σ In a further application, the queue can be used to store requests, and each priority is assigned It is not the same 'so that the request in the high priority column will be executed first, until all higher priority requests are completed, and the low priority queue request is executed. This priority based schedule = extensive Used, because the more important requests are processed first, the table will be better than the general scheduling method. However, this priority-based scheduler still has some drawbacks. One drawback is that low priority requests can hinder high priority requests. When the = and the initiator randomly issue a request, it may happen that all of the requests are executed after the execution of the low priority request. However, when a new high priority request arrives and a low priority request is still being executed, this right request must wait until the low priority request is executed before starting. = It can be seen that the high priority green request will be delayed, and the overall cylinder efficiency = 1295014, 14565twf.doc /g SUMMARY OF THE INVENTION In one aspect, the present invention provides an apparatus and method for scheduling requests sent to resource elements. The apparatus includes a high priority request queue for storing a plurality of high priority requests to the resource device; a low priority request queue for storing a low priority request to the resource element , wherein the priority order of each high priority request is higher than the priority order of the low priority request; a history counter is used to store a message related to at least one of the two adjacent heights Request interval between priority requests, and a scheduling module, sorting high priority requests and low priority requests according to the aforementioned information. On the other hand, the present invention provides a scheduling request for resource elements. The method includes receiving a plurality of high priority requests and a low priority request, wherein each of the high priority requests has a higher priority than the low priority requests; storing a message, and the information is related to At least one request interval between two adjacent high priority requests; and sorting high priority requests and low priority requests based on this informationAccording to a preferred embodiment of the present invention, wherein the information indicates that there is sufficient time for the first part of the low priority request to operate, the low priority request is split into the first part and at least the other part, and is low. The first part of the priority will be granted to request resource elements. In accordance with a preferred embodiment of the present invention, the information includes an average of a plurality of request intervals, wherein the average is obtained by weighting each request interval and then adding. 1295014 14565twf.doc/g Since the present invention considers the request interval occurring in the high priority queue and sorts the low priority request based on this information, it will cause the low priority request to be granted access to the high priority request request interval, and high Opportunities for priority requests to be blocked by low priority requests will also be reduced due to the special ordering of low priority requests. The above and other objects, features and advantages of the present invention will become more <RTIgt; [Embodiment] Please refer to FIG. 1 , which is a block diagram of a device according to a preferred embodiment of the present invention. In the embodiment, the device 10 includes a high priority queue 12 and a low priority queue 14 . A record counter 16 and a scheduling module 18. The agents 100-106 share the resources in the resource element 11 by issuing a request to the device 10, such as a dynamic random access memory (dram). This device 10 receives multiple requests and sorts the requests received > to have these agents 100-106 access the resource elements. In the present embodiment, the requests issued by the agents 100 and 102 are collected in the above-mentioned high priority queue 12, and the requests issued by the agents 1〇4 and 1〇6 are set to low priority. In column 14. The request in the high priority queue 12 t has a higher priority than the request in the low priority column 14. Further, the request in the low priority queue 14 can be granted permission to execute only after the high priority queue 12 is emptied. That is, the agents 104 and 106 have lower priority when accessing the resource element 11〇. For resource element no, the length of time used by the requests issued by the agents 1〇4 1295014, 14565twf.doc/g, and l6 must be properly planned. Data access operations that do not result in high priority agent efficiencies are inefficient. / or low priority agent In order to determine the appropriate length of the cycle, the record counts information about the request interval between two adjacent high priority requests (i.e., requests in -= 12). Special two knowledge, Ji = Lai ϋ 16 self-first _ high priority request " number has ί less clock passing, when a new high priority request arrives, '= state 16 stores the count of the request interval ^ However, Record _ 16 in addition to Α two start two times - the count. Information. For example, the average of the other request intervals described above. ^ = Separated or: This = Right: Seek,: Add the sum to get " ^ Time. x Package 3 in the description of the various information in the poor news is called 18 can be used according to the information on the existence of the record counter == First 24 requests for low priority. First, scheduling module 18 agrees to operate at high ίΓ2, while in high priority (four) u is/monthly, group 18 will start sorting at low priority. The request in column 14 is specific. When the information stored in the record counter 16 indicates that there is enough time to perform the operation of the low money request, the low priority request will be allowed to be dumped. If this message indicates that there is not enough l295〇l4, 14565twf.doc/g time to perform a complete low priority request operation, group 18 will determine if the first part of the low priority request can be granted. There is sufficient time for the first part of the low priority request to operate the Japanese, the priority request will be divided into the first part and at least the other part, and: the first part of the priority request can be granted to request the resource element. Please refer to FIG. 2, which illustrates a preferred embodiment of the present invention. A scheduler f flow chart 'This fiscal method can be used (4) - (4) component scheduling Jing ^. First 'receive high-green request (and / or) low priority request irf request (step lake) And again, storing the foregoing information, and this request is interrupted by a request interval between two adjacent high priority requests, and the received request is sorted according to the information (step s2 according to the present invention) In the case of the actual addition, in order to describe the schedule in more detail, the 'eye phase 3' is described in more detail in the step S2Q4 of Fig. 2, and the implementation of the first check is high. Priority: Column 2 is not 24 (step test). #高县剌 has not been emptied:, the two excellent riding request will be given the appropriate _ order to save (four) source Yu (step S3 〇 2). However 'if high priority When the column is emptied, the above information will be obtained (step S3〇4) and the low priority request sorting will start to operate. After obtaining the above mentioned information, the information will be checked and judged = the poor news Is the available time within the length sufficient for the low t first requested operation (step S pass). If the available time is long enough to perform a complete low priority request operation, then the low priority request will be subjected to the operation requested by the resource element by the reference 1295014 14565twf.doc/g (step S308). When the available time is not enough to perform a complete low priority request operation, it is considered to split the low priority request into several parts. For example, as shown in step S310, the low priority request may be checked first to determine A first part that is completed within the available time of the description. If the operation of any part of the low priority request cannot be supported by the day guard, the flow returns to step S300. However, if the available time is sufficient When the first part of the operation is low priority, the low priority request is divided into two parts, one of which is the first part (step S312). After splitting the low priority request, the first part of the low priority request can be granted 2 access resource elements, while the other part of the low priority request remains in the low priority 4 column 'waiting for the next grant access Operation. In the present invention, the appropriate length between the requests of the low priority agents for accessing the resource elements is determined by the evaluation values of the last and upcoming request intervals. The present invention optimizes the low priority generation performance; the efficiency of the sub-fetching, and the minimization of the pause of the high priority request can be limited to f (four) has been implemented to achieve the above, but it is not used to the scope and The skilled artisan can make some changes and refinements without departing from the spirit of the invention. Therefore, the protection of the present invention [simple description of the drawings] 々千图1 Fig. 2 ^ is a better #_- Figure. And is not a scheduling method flow of the preferred embodiment of the present invention 11 1295014 14565twf.doc/g diagram. FIG. 3 is a flow chart showing a detailed flowchart of the preferred embodiment of the present invention. [Main component symbol description] 10 · Two kinds of requests for scheduling to resource elements 12 · South priority column 14 : Low priority 彳 column 16 : Record counter 18 : Scheduling modules 100 to 106 : Agent 110: resource element S200: receiving a high bit request and a low bit request S202: determining - information, the request interval between the information is 7 arsenic, and the priority is S204. According to the information, the received request S300: whether the high priority queue is selected Is cleared S302: grant high priority requester S304: whether the elapsed time for obtaining the information is long enough to make low priority request S308: agree low priority request ^ line low priority request off: judge the available time is the first part of operation length疋 with ==: $ first request is split into two parts S3M · agree with the first part of the low priority request 12