<html xmlns:v="urn:schemas-microsoft-com:vml" xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:w="urn:schemas-microsoft-com:office:word" xmlns:m="http://schemas.microsoft.com/office/2004/12/omml" xmlns="http://www.w3.org/TR/REC-html40">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=us-ascii">
<meta name="Generator" content="Microsoft Word 15 (filtered medium)">
<style><!--
/* Font Definitions */
@font-face
{font-family:"Cambria Math";
panose-1:2 4 5 3 5 4 6 3 2 4;}
@font-face
{font-family:Calibri;
panose-1:2 15 5 2 2 2 4 3 2 4;}
/* Style Definitions */
p.MsoNormal, li.MsoNormal, div.MsoNormal
{margin:0cm;
margin-bottom:.0001pt;
font-size:11.0pt;
font-family:"Calibri",sans-serif;}
a:link, span.MsoHyperlink
{mso-style-priority:99;
color:#0563C1;
text-decoration:underline;}
a:visited, span.MsoHyperlinkFollowed
{mso-style-priority:99;
color:#954F72;
text-decoration:underline;}
p.MsoListParagraph, li.MsoListParagraph, div.MsoListParagraph
{mso-style-priority:34;
margin-top:0cm;
margin-right:0cm;
margin-bottom:0cm;
margin-left:36.0pt;
margin-bottom:.0001pt;
font-size:11.0pt;
font-family:"Calibri",sans-serif;}
span.EmailStyle17
{mso-style-type:personal-compose;
font-family:"Calibri",sans-serif;
color:windowtext;}
.MsoChpDefault
{mso-style-type:export-only;
font-family:"Calibri",sans-serif;}
@page WordSection1
{size:612.0pt 792.0pt;
margin:72.0pt 90.0pt 72.0pt 90.0pt;}
div.WordSection1
{page:WordSection1;}
--></style><!--[if gte mso 9]><xml>
<o:shapedefaults v:ext="edit" spidmax="1026" />
</xml><![endif]--><!--[if gte mso 9]><xml>
<o:shapelayout v:ext="edit">
<o:idmap v:ext="edit" data="1" />
</o:shapelayout></xml><![endif]-->
</head>
<body lang="EN-US" link="#0563C1" vlink="#954F72">
<div class="WordSection1">
<p class="MsoNormal">Sagi Grimberg presented an RFC for supporting RDMA erasure coding offloads.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">RAID schemes are common in networked storage solutions, such as Ceph and Gluster.<o:p></o:p></p>
<p class="MsoNormal">Erasure coding is a generalization of different RAID schemes, such as mirroring, parity block, and dual-parity blocks.<o:p></o:p></p>
<p class="MsoNormal">Specifically, using erasure coding, a system can withstand multiple disk failures by adding redundancy code blocks to the original data.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">In a distributed setting, different data and redundancy blocks typically reside on different nodes to minimize correlated failures.<o:p></o:p></p>
<p class="MsoNormal">Thus, the task of calculation of the code blocks, which is highly cpu intensive, is related to data transfer.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">The proposed API offers support for the following use cases:<o:p></o:p></p>
<p class="MsoNormal">(1) Synchronous encoding (for generating code blocks)<o:p></o:p></p>
<p class="MsoNormal">(2) Synchronous decoding (for reconstructing data following a disk loss)<o:p></o:p></p>
<p class="MsoNormal">(3) Asynchronous encoding<o:p></o:p></p>
<p class="MsoNormal">(4) Asynchronous decoding<o:p></o:p></p>
<p class="MsoNormal">(5) Encode + data transfer<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">The last use-case is particularly interesting because it enables the Verbs provider to encode the data and transfer the outcome to remote peers without any SW synchronization.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">The API consists of an erasure coding (EC) context, which is initialized with the code parameters and calculation matrix.<o:p></o:p></p>
<p class="MsoNormal">This context may be used for multiple encoding/decoding operations.<o:p></o:p></p>
<p class="MsoNormal">When operations are invoked, the caller providers a memory layout specifying the scatter/gather entries that constitute the data and code blocks, as well as the block size, for the specific operations.<o:p></o:p></p>
<p class="MsoNormal">Asynchronous operations are also given a callback function to indicate completion.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Finally, the encode+transfer primitive also requires a vector of strip objects, one for each data or code block.<o:p></o:p></p>
<p class="MsoNormal">The stripe objects specify the QP and RDMA operation to conduct following encoding completion.<o:p></o:p></p>
<p class="MsoNormal">Multiple operations may be posted concurrently on the same EC context, allowing, for example, the pipelining of multiple disk writes.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">In terms of ordering, all operations submitted to the same EC context are executed in submission order.<o:p></o:p></p>
<p class="MsoNormal">However, in encode+transfer operations, the ordering between transfers to different peers is not guaranteed.<o:p></o:p></p>
<p class="MsoNormal">--Liran<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
</div>
</body>
</html>