完整後設資料紀錄
DC 欄位語言
dc.contributor.author戴函昱en_US
dc.contributor.authorHan-Yu Taien_US
dc.contributor.author張瑞川en_US
dc.contributor.authorRuei-Chuan Changen_US
dc.date.accessioned2014-12-12T03:10:13Z-
dc.date.available2014-12-12T03:10:13Z-
dc.date.issued2006en_US
dc.identifier.urihttp://140.113.39.130/cdrfb3/record/nctu/#GT009455612en_US
dc.identifier.urihttp://hdl.handle.net/11536/82131-
dc.description.abstract在原本的Xen網路架構中,一個客虛擬機器要必須要透過主虛擬機器才能夠存取網路。這個架構的好處是可以重用主虛擬機器中的作業系統的驅動程式,而不需要另外開發。但是這個架構也造成了額外的負擔,因為主虛擬機器和客虛擬機器之間必須互相配合溝通來完成網路的存取,因此使得客虛擬機器的網路吞吐量降低。 在這篇論文中,我們提出一個driver-domain-bypassing network (DBNet) 架構,藉此降低主客虛擬機器之間溝通所造成的額外負擔,並且提高客虛擬機器的網路吞吐量。我們將網路卡驅動程式從主虛擬機器中搬移到virtual machine monitor (VMM)中,但是為了避免複雜化VMM,我們只搬移驅動程式中和效能最相關的部分,其餘的部分保留在主虛擬機器中。另外,為了讓VMM可以正確地將封包發送到遠端機器或是客虛擬機器,並且避免複雜化VMM,我們在VMM中實做了一個簡化過的bridge。最後,為了保護VMM不會受到驅動程式錯誤的影響而造成當機,我們將網路卡驅動程式放在一個獨立的segment中,並且降低網路卡驅動程式的執行權限,藉此保護VMM。 在測量效能的實驗中顯示,DBNet架構可以有效的提升客虛擬機器的網路效能。在只有一個客虛擬機器的環境下,且在傳輸封包的測試中幾乎可以達到和Linux相同的效能,而在接收封包的測試中則可以用較少的CPU utilization達到更多的網路吞吐量。在多個客虛擬機器的環境下,不論是傳送或是接收封包的測試中,DBNet都可以有效的達到比原本Xen網路架構更好的效能。zh_TW
dc.description.abstractIn the original Xen network architecture, each guest virtual machine has to access the network through a host virtual machine. The most advantage of this architecture is to reuse device drivers within the operating system running within the host virtual machine. However, this architecture causes lots of additional overhead to perform communication between the guest virtual machine and the host machine while reduces the network throughput. In this thesis, we propose a driver-domain-bypassing network (DBNet) architecture to reduce the communication overhead and increase the network throughput of a guest virtual machine by bypassing the host virtual machine. We migrate a NIC driver from the host virtual machine into the virtual machine monitor (VMM). In order to avoid complicating the VMM, we only move the performance-critical part of the NIC driver into the VMM and keep other parts in the host virtual machine. Moreover, for allowing the VMM can dispatch a packet to a target guest domain or a remote machine correctly, and avoiding complicating the VMM, we implement a simplified bridge in the VMM. Finally, to protect the VMM from crashing by driver faults, we put the NIC driver into an independent driver segment and lower its privilege level. Our performance measure shows that in the one-guest-domain environment, the DBNet architecture can achieve nearly the same performance with Linux in TCP transmission test, and achieve higher network throughput while use less CPU utilization in TCP reception test. Beside, DBNet architecture can also achieve better performance for each guest virtual machine than the original Xen network architecture in the multi-guest-domain environment.en_US
dc.language.isozh_TWen_US
dc.subject虛擬機器zh_TW
dc.subject驅動程式zh_TW
dc.subject網路zh_TW
dc.subjectvirtual machineen_US
dc.subjectdevice driveren_US
dc.subjectnetworken_US
dc.subjectprotectionen_US
dc.title藉由繞過驅動程式域改善Xen虛擬機器之網路吞吐量zh_TW
dc.titleImproving Xen Network Throughput by Bypassing Driver Domainen_US
dc.typeThesisen_US
dc.contributor.department資訊科學與工程研究所zh_TW
顯示於類別:畢業論文


文件中的檔案:

  1. 561201.pdf

若為 zip 檔案,請下載檔案解壓縮後,用瀏覽器開啟資料夾中的 index.html 瀏覽全文。