通过写这些乱七八糟,走了好多弯路:
- 列出具体问题
- 问题相关的reference
- 有无参考
所以今天解决的问题是:
- routing的目标节点是server node还是cache node
答:cache node. 因为本来想优化的就是,在cache node也存的情况下,存在从remote cache node 拉取信息的延时
大于server node的情况。 - 若为cache node,delivery 变量是hop by hop 的变量还是path 统称(忽略hop by hop)?
答:path 统称。不然可能需要重写CP - delivry strategy应为[0,1] or {0,1}?
答:delivery={0,1};cost定义为Jian-Li形式。假定,每个节点都用stability score s(k)来capture the feature of node's resilience, and function u(s(k))来表示如果路径中的缓存节点k拓机产生的消耗。节点k越不稳定,u越大,cost越大。当在当前路径中的所有节点全都不稳定时,产生的cost可能比server node下载的cost要大。 - 在delivery 的过程中,途径每个节点CP设定是否需要考虑?
[参考:+1
DR-Cache : Distributed Resilient Caching with Latency Guarantees:each user can also access a subset of file in-network caches. refer a connection between a uer and a cache as a cache pat."我们可以引用此说法,user到in-netowrk cache的路径成为cache path;+1
Wei Jiang:“we assme that the MBS has a central control unit that has the location information and the caching information of all UEs and FBSs. It is convenient for MBS to allocate resources for all communication links and deply certain content items at UEs and FBSs(这里的transimission policy 只是选择是FBS还是UE传输。)”;-1
FemtoCaching: Wireless content delivery through distributed caching helpers:论文中将用户到每个helper的delay进行有小到大排序,延时第j大的helper存则小于该延时的helper不缓存,为啥要排序?+1
Information Resilience through User-Assisted Caching in Disruptive Content-Centric Networks:router转发。]。
为了最大hit rate考虑节点间的协作,而节点间的协作需要路由转发到目的节点,从不同的目的节点拉取信息由于路径不同,产生的delay不同,因此当delay超过server时,如何分配delivery从而优化QoE.}$ - CP与DS的组合?
是在CP一定的情况下,选择哪条路径?还是到达缓存节点路径的cost影响CP的缓存?[参考 "FemtoCaching : Wireless Video Content Delivery through Distributed Caching Helpers,Golrezaei, Negin"作者用sum表示the probability that usre k requests one of the files that are accessible through its local helpers 而不是连乘]
1. Cooperation 及 Model
- 源由
忘了要说啥,哈哈哈哈哈哈哈哈
重新想吧
cooperation:就意味着CP(i)=0,其他至少一个CP(j)=1;从而衍生出连乘。
今天想了好久,在delivery model中是否需要cooperation。
实际上是不需要的。
证明:引用下文1。
If user i requests file f and it is present in tha cache, then the request is served immediately. We refer to this event as a cache hit. However, if content j is not present in the cache, the cache then forwards the request to the back-end server. Note that in case of a cache miss, the cache can decide whether to keep the download content.
综上,这里并未涉及到local cache 未满足,转而其他的同级的cache node。
模型的mitivation:
- 优化变量: CP && Delivery Routing
本来考虑的场景是:用户 i 产生请求,local cache先反应,如果local cache generate cache miss, 请求被转向其他同级cache node。 而a request routed along a path fails if one of the intermediate nodes on the path is broken.
we call this process delivery cost. The more resilient the intermediate nodes are , the smaller the delivery cost is. 在同级协作过程中,如果产生的delivery cost>server cost,delivery cost 则不是一个很好的cache 选择。We seem to take cooperation into consideration in the analysis process. 基于上述描述,建模过程中,需要考虑CP(local cache),&&delivery strategy*CP(cooperation cache nodes),变量>=3,不会解。呵呵呵呵呵呵呵
问题梳理:
通常为max hit rate,选择cooperation,但从remote cache node拉取数据的是时候,由于节点的不稳定性(如拓机)及转发时的消耗。Failure and caching performance are inherently tired together in cache networks since failures result in further searching and routing which, in turn, can significantly increase latency. 路径越长,请求被成功routed to cache node的概率越小,系统产生的cost越大。假定route to cache node过程中生成的cost与系统选择路径概率成反比。或者cost越小,选择在CP的情况下,如何选择delivery,使系统cost最小?
问题:
因为本文中的核心是从某些remote cache nodes拉取信息的时候,由于节点的resilience,其产生的cost/delay可能要大于server node,所以要design delivery strategy。即:How to express it? {relevant:
delivery strategy的定义值不同影响cost的定义}
option1:
delivery strategy=[0,1]-->路径消耗的cost与选择路径的概率成反比?
分析:这样定义capture the delivery distribution.
[Reference "Jointly"这样定义:
In the case of randomized routing, variables r are random. We randomize routing by allowing requests to be routed over a random paht in P( P is set of paths of routing reuqest for file i to server node) ]
option2:
delivery strategy={0,1}-->将cost定义成Jian-Li形式?
分析:定义成整数形式,存在的问题是否需要将传输策略变成至少有一个能满足用户需要。[所以有了后续的cooperation分析的想法]。应该不需要。如果没有deliverycelue,至少有一个CP满足需要考虑(如异质分层网络),consider reference[on the complexity of optimal request routing and content caching in heterogeneous cache network && Wei jiang]至少一个的策略不需要考虑,直接routing指向固定的cache node。
进一步的,我们做如下抽象。
每个用户请求通过一定的路径到达cache node。一个请求一次只能由一个cache node响应(但可存在多个能满足请求的cache nodes)。 假设用户i产生对文件f的请求,请求首先到local cache node,如果local cache node没有cache content, 则请求被routing到其他remote cache node。 we assume that requests to content are routed along paths in G, G中包含对不同文件请求的路径P(f,i,n)。选择每种对文件f请求的路径的概率服从分布Q(f,i,n)。每条路径的weight值为cost(u,c,f)。且cost正比于(1-q)。cache node满足用户的条件是: CP && delivery Strategy(DS)。每个用户一次只能被一个cache node所服务。
上述证明。即local cost和delivery cost通看成一类,不做区分。我们假设cost正比于(1-delivery strategy)
- 目标: 最小化系统cost
- 约束:
- 框架
-
《On the Complexity of Optimal Request Routing and Content Caching in Heterogeneous Cache Networks》
-
《Inter-cluster Copperation for wireless D2D Caching Networks》
-《Proactive Caching Strategies in Heterogeneous Networks With Device-to-Device Communications》
- 总结
2. Jointly routing
- DR-Cache: Distributed Resulient Caching with Latency Guarantees(Jian Li)
- Jointly Optimal Routing and Caching for Arbitrary Network Topologies(Stratis Ioannidis)
- 共同点
场景上:
① 请求从源节点到server node,至少有一个cache node能满足用户请求。通过min全反来表达。
min Routing*multiply(1-x(cache node in routing)) - 不同点
优化变量不同,影响变量的因素不同。
网友评论