[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82704-en":3,"doc-seo-82704-105":30,"detail-sidebar-cat-0-en-105":90},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},82704,4810365810221,"Aurora","https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d",8,"Research & Report","Cross-IP Request Coalescing: Relocating the Fan-out Point in Virtualized I/O","Cloud data centers use virtualization to serve AI workloads in multi-tenant settings, where storage I/O through the virtualization layer becomes a key performance limiter. A single user request may traverse multiple IP blocks, causing functional units to fan out into multiple stack traversals, accumulating round-trip and context-switching overhead. This letter pinpoints per-device submission structure as a dominant source, even under user-space kernel-bypass frameworks like SPDK. It proposes cross-IP request coalescing that relocates the fan-out point to the backend, reducing N device requests to share one traversal and achieving up to 1.78× lower latency as concurrency rises.","Cross-IP Request Coalescing: Relocating the Fan-out Point in Virtualized I/O  \nKiseok Kim, Hyeontae Joo, and Hwangnam Kim, Member, IEEE  \narXiv :2607 .02969v 1 [ cs .DC] 3 Jul 2026  \nAbstract—Cloud data centers rely on virtualization technologies to serve AI workloads in multi-tenant environments. With the growing scale of data-intensive AI workloads, the performance of storage I/O paths at the virtualization layer has become a critical factor. A single user request often crosses multiple IP blocks; functional units such as storage, GPU, and accelerator devices, under virtualization, fanning out into separate stack traversals between the guest and the backend. As a result, round-trip and context-switching overheads accumulate with the number of devices. In this letter, we identify that a dominant factor in this overhead lies not in the kernel-mediated I/O path alone, but in the per-device submission structure itself, which persists even in user-space, kernel-bypass storage frameworks such as SPDK. To address this, we propose cross-IP request coalescing, which relocates the fan-out point from the guest to the backend. As a first instantiation, we apply this principle to block storage I/O. The guest submits a single coalesced request, and an SPDK acceleration block device (bdev) decomposes it and dispatches per-device I/O locally, so that N device requests share a single stack traversal instead of N. Evaluation in an SPDKbased virtualized environment shows that the proposed approach achieves up to 1.78× lower latency than the per-device baseline, with the benefit growing as concurrency increases.  \nIndex Terms—Virtualization, SPDK, vhost-user, storage I/O, multi-device dispatch, data center workloads.  \nI. INTRODUCTION  \nVIRTUALIZATION technologies, ranging from  \nhypervisor-based virtual machines to container-based platforms, support cloud and AI infrastructure, yet their I/O paths incur overhead across multiple layers as workloads demand concurrent access to high-speed storage devices [1] . The rapid growth of data-intensive AI workloads also leads to concurrent requests across multiple devices [2] . As GPUs, memory, and storage scale up, the path overhead of data transfer has emerged as a critical issue in fully utilizing hardware performance. Among these issues, device I/O performance is one of the performance-critical paths, yet a residual inefficiency persists even in existing solutions [3] . Existing techniques such as user-space I/O acceleration, hardware-assisted offloading, and device assignment reduce software overhead along the I/O path. However, they do not directly address the per-device submission overhead that scales  \nManuscript received November 00, 2000; revised February 00, 2000 .(Corresponding author: Hwangnam Kim)  \nKiseok Kim and Hyeontae Joo contributed equally to this work.  \nKiseok Kim, Hyeontae Joo, and Hwangnam Kim are with the Korea University, Seoul 02841, Republic of Korea (e-mail: {kisuk528, motern800, and [hnkim](hnkim}@korea.ac.kr)[}](hnkim}@korea.ac.kr)[@korea.ac.kr](hnkim}@korea.ac.kr))  \nFig. 1. Multi-device I/O fanning out across the virtualization stack.  \nwith the number of devices a single request spans, as in replicated storage writes or multi-volume parallel writes, where a guest issues concurrent I/O across multiple devices [4] .  \nIn this context, we analyze the structural overhead of the virtualized multi-device I/O path on QEMU, a representative type-2 hypervisor, and observe that repeated per-device traversals occur between the guest and the backend under concurrent multi-device access, as illustrated in Fig. 1. We further observe that this overhead persists even with an SPDKvhost-user backend. Although user-space I/O stacks such as SPDK bypass the kernel path and provide benefits in highspeed storage I/O [5], they leave the per-request submission structure intact, so the repetitive guest-to-backend overhead persists regardless of kernel bypass. Inspired by memory coalescing, which","cbCaipNyiRZ9o3aD","https://ap.wps.com/l/cbCaipNyiRZ9o3aD","pdf",702294,3,1,4,"English","en",105,"# Introduction\n# Background and Motivation\n## Virtualization I/O path in QEMU","[{\"question\":\"What problem does cross-IP request coalescing target in virtualized I/O?\",\"answer\":\"It targets the overhead caused by per-device submission structure when a single request fans out across multiple devices through the virtualization stack.\"},{\"question\":\"Why does the overhead persist even with user-space kernel-bypass storage frameworks like SPDK?\",\"answer\":\"User-space I/O stacks such as SPDK bypass the kernel path but keep the per-request submission structure intact, so repetitive guest-to-backend overhead still occurs.\"},{\"question\":\"How does the proposed approach reduce latency for multi-device block storage I/O?\",\"answer\":\"The guest submits one coalesced request, and an SPDK bdev decomposes and dispatches per-device I/O locally, letting N device requests share a single stack traversal instead of N.\"}]",1784182401,10,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":28},"cross-ip-request-coalescing-relocating-the-fan-out-point-in-virtualized-io","",{"@graph":36,"@context":84},[37,52,67],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":20},"https://docshare.wps.com/document/research-report/",{"item":51,"name":13,"@type":43,"position":22},"https://docshare.wps.com/document/cross-ip-request-coalescing-relocating-the-fan-out-point-in-virtualized-io/82704/",{"url":51,"name":13,"@type":53,"author":54,"headline":13,"publisher":56,"fileFormat":59,"inLanguage":24,"description":14,"dateModified":60,"datePublished":61,"encodingFormat":59,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":55},"Person",{"url":41,"name":57,"@type":58},"DocShare","Organization","application/pdf","2026-07-24","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":20},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"What problem does cross-IP request coalescing target in virtualized I/O?","Question",{"text":74,"@type":75},"It targets the overhead caused by per-device submission structure when a single request fans out across multiple devices through the virtualization stack.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"Why does the overhead persist even with user-space kernel-bypass storage frameworks like SPDK?",{"text":79,"@type":75},"User-space I/O stacks such as SPDK bypass the kernel path but keep the per-request submission structure intact, so repetitive guest-to-backend overhead still occurs.",{"name":81,"@type":72,"acceptedAnswer":82},"How does the proposed approach reduce latency for multi-device block storage I/O?",{"text":83,"@type":75},"The guest submits one coalesced request, and an SPDK bdev decomposes and dispatches per-device I/O locally, letting N device requests share a single stack traversal instead of N.","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":57,"og:description":14},"article",{"robots":88,"canonical":51},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,119,122,127,130,133],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":22,"doc_module":4,"doc_module_name":46,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":46,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":46,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":46,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":46,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":46,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":29,"doc_module":4,"doc_module_name":46,"category_name":131,"show_sort_weight":29,"slug":132},"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":46,"category_name":135,"show_sort_weight":105,"slug":136},19,"General","general"]