diff --git a/css/main.css b/css/main.css
index 74094ad..b4457c9 100644
--- a/css/main.css
+++ b/css/main.css
@@ -2729,7 +2729,7 @@ hr {
   vertical-align: middle;
 }
 .links-of-author a::before {
-  background: #e847ff;
+  background: #75fff6;
   display: inline-block;
   margin-right: 3px;
   transform: translateY(-2px);
diff --git a/publications/index.html b/publications/index.html
index a56c87b..8251fe5 100644
--- a/publications/index.html
+++ b/publications/index.html
@@ -132,7 +132,7 @@ <h1 class="post-title" itemprop="name headline">Publications
       
       
       <div class="post-body">
-          <!-- begin-publist-4c41a19c -->
+          <!-- begin-publist-c442244d -->
 <style type="text/css">
 :root{--color-accent-emphasis:#0969da;--color-accent-fg:#0969da;--color-attention-emphasis:#bf8700;--color-attention-fg:#9a6700;--color-border-default:#d0d7de;--color-border-muted:#d8dee4;--color-border-subtle:rgba(27,31,36,0.15);--color-canvas-overlay:#fff;--color-canvas-subtle:#f6f8fa;--color-danger-emphasis:#cf222e;--color-danger-fg:#cf222e;--color-diffstat-addition-bg:#2da44e;--color-done-emphasis:#8250df;--color-fg-default:#24292f;--color-fg-muted:#57606a;--color-fg-on-emphasis:#fff;--color-neutral-emphasis:#6e7781;--color-neutral-emphasis-plus:#24292f;--color-neutral-muted:rgba(175,184,193,0.2);--color-neutral-subtle:rgba(234,238,242,0.5);--color-primer-canvas-backdrop:rgba(27,31,36,0.5);--color-select-menu-backdrop-border:transparent;--color-select-menu-tap-focus-bg:#b6e3ff;--color-select-menu-tap-highlight:rgba(175,184,193,0.5);--color-shadow-large:0 8px 24px rgba(140,149,159,0.2);--color-success-emphasis:#2da44e;--color-success-fg:#1a7f37}.tooltipped{position:relative}.tooltipped:after{-webkit-font-smoothing:subpixel-antialiased;word-wrap:break-word;background:var(--color-neutral-emphasis-plus);border-radius:6px;color:var(--color-fg-on-emphasis);content:attr(aria-label);font:normal normal 11px/1.5 -apple-system,BlinkMacSystemFont,Segoe UI,Helvetica,Arial,sans-serif,Apple Color Emoji,Segoe UI Emoji;letter-spacing:normal;padding:.5em .75em;text-align:center;text-decoration:none;text-shadow:none;text-transform:none;white-space:pre;z-index:1000000}.tooltipped:after,.tooltipped:before{display:none;opacity:0;pointer-events:none;position:absolute}.tooltipped:before{border:6px solid transparent;color:var(--color-neutral-emphasis-plus);content:"";height:0;width:0;z-index:1000001}@keyframes tooltip-appear{0%{opacity:0}to{opacity:1}}.tooltipped:active:after,.tooltipped:active:before,.tooltipped:focus:after,.tooltipped:focus:before,.tooltipped:hover:after,.tooltipped:hover:before{animation-delay:.4s;animation-duration:.1s;animation-fill-mode:forwards;animation-name:tooltip-appear;animation-timing-function:ease-in;display:inline-block;text-decoration:none}.tooltipped-no-delay:active:after,.tooltipped-no-delay:active:before,.tooltipped-no-delay:focus:after,.tooltipped-no-delay:focus:before,.tooltipped-no-delay:hover:after,.tooltipped-no-delay:hover:before{animation-delay:0s}.tooltipped-multiline:active:after,.tooltipped-multiline:focus:after,.tooltipped-multiline:hover:after{display:table-cell}.tooltipped-s:after,.tooltipped-se:after,.tooltipped-sw:after{margin-top:6px;right:50%;top:100%}.tooltipped-s:before,.tooltipped-se:before,.tooltipped-sw:before{border-bottom-color:var(--color-neutral-emphasis-plus);bottom:-7px;margin-right:-6px;right:50%;top:auto}.tooltipped-se:after{left:50%;margin-left:-16px;right:auto}.tooltipped-sw:after{margin-right:-16px}.tooltipped-n:after,.tooltipped-ne:after,.tooltipped-nw:after{bottom:100%;margin-bottom:6px;right:50%}.tooltipped-n:before,.tooltipped-ne:before,.tooltipped-nw:before{border-top-color:var(--color-neutral-emphasis-plus);bottom:auto;margin-right:-6px;right:50%;top:-7px}.tooltipped-ne:after{left:50%;margin-left:-16px;right:auto}.tooltipped-nw:after{margin-right:-16px}.tooltipped-n:after,.tooltipped-s:after{transform:translateX(50%)}.tooltipped-w:after{bottom:50%;margin-right:6px;right:100%;transform:translateY(50%)}.tooltipped-w:before{border-left-color:var(--color-neutral-emphasis-plus);bottom:50%;left:-7px;margin-top:-6px;top:50%}.tooltipped-e:after{bottom:50%;left:100%;margin-left:6px;transform:translateY(50%)}.tooltipped-e:before{border-right-color:var(--color-neutral-emphasis-plus);bottom:50%;margin-top:-6px;right:-7px;top:50%}.tooltipped-align-right-1:after,.tooltipped-align-right-2:after{margin-right:0;right:0}.tooltipped-align-right-1:before{right:10px}.tooltipped-align-right-2:before{right:15px}.tooltipped-align-left-1:after,.tooltipped-align-left-2:after{left:0;margin-left:0}.tooltipped-align-left-1:before{left:5px}.tooltipped-align-left-2:before{left:10px}.tooltipped-multiline:after{word-wrap:break-word;border-collapse:separate;max-width:250px;white-space:pre-line;width:max-content}.tooltipped-multiline.tooltipped-n:after,.tooltipped-multiline.tooltipped-s:after{left:50%;right:auto;transform:translateX(-50%)}.tooltipped-multiline.tooltipped-e:after,.tooltipped-multiline.tooltipped-w:after{right:100%}@media screen and (min-width:0\0){.tooltipped-multiline:after{width:250px}}.tooltipped-sticky:after,.tooltipped-sticky:before{display:inline-block}.tooltipped-sticky.tooltipped-multiline:after{display:table-cell}.SelectMenu{bottom:0;display:flex;flex-direction:column;left:0;padding:16px;pointer-events:none;position:fixed;right:0;top:0;z-index:99}@media(min-width:544px){.SelectMenu{bottom:auto;left:auto;padding:0;position:absolute;right:auto;top:auto}}.SelectMenu:before{background-color:var(--color-primer-canvas-backdrop);bottom:0;content:"";left:0;pointer-events:none;position:absolute;right:0;top:0}@media(min-width:544px){.SelectMenu:before{display:none}}.SelectMenu-modal{animation:SelectMenu-modal-animation .12s cubic-bezier(0,.1,.1,1) backwards;background-color:var(--color-canvas-overlay);border:1px solid var(--color-select-menu-backdrop-border);border-radius:12px;box-shadow:var(--color-shadow-large);display:flex;flex-direction:column;margin:auto 0;max-height:66%;overflow:hidden;pointer-events:auto;position:relative;z-index:99}@keyframes SelectMenu-modal-animation{0%{opacity:0;transform:scale(.9)}}@keyframes SelectMenu-modal-animation--sm{0%{opacity:0;transform:translateY(-16px)}}@media(min-width:544px){.SelectMenu-modal{animation-name:SelectMenu-modal-animation--sm;border-color:var(--color-border-default);border-radius:6px;box-shadow:var(--color-shadow-large);font-size:12px;height:auto;margin:8px 0 16px;max-height:480px;width:300px}}.SelectMenu-header{align-items:center;border-bottom:1px solid var(--color-border-muted);display:flex;flex:none;padding:16px}@media(min-width:544px){.SelectMenu-header{padding:7px 7px 7px 16px}}.SelectMenu-title{flex:1;font-size:14px;font-weight:600}@media(min-width:544px){.SelectMenu-title{font-size:inherit}}.SelectMenu-closeButton{background-color:transparent;border:0;color:var(--color-fg-muted);line-height:1;margin:-16px;padding:16px}@media(min-width:544px){.SelectMenu-closeButton{margin:-8px -7px;padding:8px}}.SelectMenu-filter{border-bottom:1px solid var(--color-border-muted);margin:0;padding:16px}@media(min-width:544px){.SelectMenu-filter{padding:8px}}.SelectMenu-input{display:block;width:100%}@media(min-width:544px){.SelectMenu-input{font-size:14px}}.SelectMenu-list{-webkit-overflow-scrolling:touch;flex:auto;margin:0 0 -1px;overflow-x:hidden;overflow-y:auto;padding:0;position:relative}.SelectMenu-item,.SelectMenu-list{background-color:var(--color-canvas-overlay)}.SelectMenu-item{align-items:center;border:0;border-bottom:1px solid var(--color-border-muted);color:var(--color-fg-default);cursor:pointer;display:flex;overflow:hidden;padding:16px;text-align:left;width:100%}@media(min-width:544px){.SelectMenu-item{padding-bottom:7px;padding-top:7px}}.SelectMenu-list--borderless .SelectMenu-item{border-bottom:0}.SelectMenu-icon{flex-shrink:0;margin-right:8px;width:16px}.SelectMenu-icon--check{transform:scale(0);transition:transform .12s cubic-bezier(.5,.1,1,.5),visibility 0s linear .12s;visibility:hidden}.SelectMenu-tabs{-webkit-overflow-scrolling:touch;box-shadow:inset 0 -1px 0 var(--color-border-muted);display:flex;flex-shrink:0;overflow-x:auto;overflow-y:hidden}.SelectMenu-tabs::-webkit-scrollbar{display:none}@media(min-width:544px){.SelectMenu-tabs{padding:8px 8px 0}}.SelectMenu-tab{background-color:transparent;border:0;box-shadow:inset 0 -1px 0 var(--color-border-muted);color:var(--color-fg-muted);flex:1;font-size:12px;font-weight:500;padding:8px 16px;text-align:center}@media(min-width:544px){.SelectMenu-tab{border:solid transparent;border-top-left-radius:6px;border-top-right-radius:6px;border-width:1px 1px 0;flex:none;padding:4px 16px}}.SelectMenu-tab[aria-selected=true]{background-color:var(--color-canvas-overlay);box-shadow:0 0 0 1px var(--color-border-muted);color:var(--color-fg-default);cursor:default;z-index:1}@media(min-width:544px){.SelectMenu-tab[aria-selected=true]{border-color:var(--color-border-muted);box-shadow:none}}.SelectMenu-message{background-color:var(--color-canvas-overlay);border-bottom:1px solid var(--color-border-muted);padding:7px 16px;text-align:center}.SelectMenu-blankslate,.SelectMenu-loading{background-color:var(--color-canvas-overlay);padding:24px 16px;text-align:center}.SelectMenu-divider{background-color:var(--color-canvas-subtle);border-bottom:1px solid var(--color-border-muted);color:var(--color-fg-muted);font-size:12px;font-weight:500;margin:0;padding:4px 16px}.SelectMenu-list--borderless .SelectMenu-divider{border-top:1px solid var(--color-border-muted)}.SelectMenu-list--borderless .SelectMenu-divider:empty{border-top:0;padding:0}.SelectMenu-footer{border-top:1px solid var(--color-border-muted);color:var(--color-fg-muted);font-size:12px;padding:8px 16px;text-align:center;z-index:0}@media(min-width:544px){.SelectMenu-footer{padding:7px 16px}}.SelectMenu--hasFilter .SelectMenu-modal{height:80%;margin-top:0;max-height:none}@media(min-width:544px){.SelectMenu--hasFilter .SelectMenu-modal{height:auto;margin-top:8px;max-height:480px}}.SelectMenu-closeButton:focus,.SelectMenu-item:focus,.SelectMenu-tab:focus{outline:0}.SelectMenu-item:hover{text-decoration:none}.SelectMenu-item[aria-checked=true]{color:var(--color-fg-default);font-weight:500}.SelectMenu-item[aria-checked=true] .SelectMenu-icon--check{transform:scale(1);transition:transform .12s cubic-bezier(0,0,.2,1),visibility 0s linear;visibility:visible}.SelectMenu-item:disabled,.SelectMenu-item[aria-disabled=true]{color:var(--color-fg-muted);pointer-events:none}@media(hover:hover){.SelectMenu-closeButton:hover,body:not(.intent-mouse) .SelectMenu-closeButton:focus{color:var(--color-fg-default)}.SelectMenu-closeButton:active{color:var(--color-fg-muted)}.SelectMenu-item:hover,body:not(.intent-mouse) .SelectMenu-item:focus{background-color:var(--color-neutral-subtle)}.SelectMenu-item:active{background-color:var(--color-canvas-subtle)}body:not(.intent-mouse) .SelectMenu-tab:focus{background-color:var(--color-select-menu-tap-focus-bg)}.SelectMenu-tab:hover{color:var(--color-fg-default)}.SelectMenu-tab:not([aria-selected=true]):active{background-color:var(--color-canvas-subtle);color:var(--color-fg-default)}}@media(hover:none){.SelectMenu-item:active,.SelectMenu-item:focus{background-color:var(--color-canvas-subtle)}.SelectMenu-item{-webkit-tap-highlight-color:var(--color-select-menu-tap-highlight)}}.IssueLabel{border:1px solid transparent;border-radius:2em;display:inline-block;font-size:12px;font-weight:500;line-height:18px;padding:0 7px}.IssueLabel .g-emoji{display:inline-block;font-size:1em;line-height:1;position:relative;top:-.05em}.IssueLabel:hover{text-decoration:none}.IssueLabel--big{line-height:22px;padding-left:10px;padding-right:10px}.labels{position:relative}.label,.Label{background-color:transparent!important;border:1px solid transparent;border-color:var(--color-border-default);border-radius:2em;display:inline-block;font-size:12px;font-weight:500;line-height:18px;padding:0 7px}.label:hover,.Label:hover{text-decoration:none}.Label--large{line-height:22px;padding-left:10px;padding-right:10px}.Label--inline{display:inline;font-size:.9em;padding:.1667em .5em}.Label--primary{border-color:var(--color-neutral-emphasis);color:var(--color-fg-default)}.Label--secondary{border-color:var(--color-border-default);color:var(--color-fg-muted)}.Label--info{border-color:var(--color-accent-emphasis);color:var(--color-accent-fg)}.Label--success{border-color:var(--color-success-emphasis);color:var(--color-success-fg)}.Label--warning{border-color:var(--color-attention-emphasis);color:var(--color-attention-fg)}.Label--danger{border-color:var(--color-danger-emphasis);color:var(--color-danger-fg)}.state,.State{border-radius:2em;display:inline-block;font-size:14px;font-weight:500;line-height:20px;padding:5px 12px;text-align:center;white-space:nowrap}.state,.State,.State--draft{background-color:var(--color-neutral-emphasis);border:1px solid transparent;color:var(--color-fg-on-emphasis)}.State--open{background-color:var(--color-success-emphasis)}.State--merged,.State--open{color:var(--color-fg-on-emphasis)}.State--merged{background-color:var(--color-done-emphasis)}.State--closed{background-color:var(--color-danger-emphasis);color:var(--color-fg-on-emphasis)}.State--small{font-size:12px;line-height:24px;padding:0 10px}.State--small .octicon{width:1em}.Counter{background-color:var(--color-neutral-muted);border:1px solid transparent;border-radius:2em;color:var(--color-fg-default);display:inline-block;font-size:12px;font-weight:500;line-height:18px;min-width:20px;padding:0 6px;text-align:center}.Counter:empty{display:none}.Counter .octicon{opacity:.8;vertical-align:text-top}.Counter--primary{background-color:var(--color-neutral-emphasis);color:var(--color-fg-on-emphasis)}.Counter--secondary{background-color:var(--color-neutral-subtle);color:var(--color-fg-muted)}.diffstat{color:var(--color-fg-muted);cursor:default;font-size:12px;font-weight:600;white-space:nowrap}.diffstat-block-added,.diffstat-block-deleted,.diffstat-block-neutral{display:inline-block;height:8px;margin-left:1px;outline-offset:-1px;width:8px}.diffstat-block-deleted{background-color:var(--color-danger-emphasis);outline:1px solid var(--color-border-subtle)}.diffstat-block-added{background-color:var(--color-diffstat-addition-bg);outline:1px solid var(--color-border-subtle)}.diffstat-block-neutral{background-color:var(--color-neutral-muted);outline:1px solid var(--color-border-subtle)}.details-overlay[open]>summary:before{background:transparent;bottom:0;content:" ";cursor:default;display:block;left:0;position:fixed;right:0;top:0;z-index:80}.details-overlay-dark[open]>summary:before{background:var(--color-primer-canvas-backdrop);z-index:99}.details-reset>summary{list-style:none}.details-reset>summary:before{display:none}.details-reset>summary::-webkit-details-marker{display:none}.post-header .post-meta{display:none}.posts-expand{padding-top:0!important}.pub-list{line-height:1.5}.pub-list section.year{align-items:flex-start;display:flex;margin-left:3.5em;padding:1em 0 0;position:relative}.pub-list section.year ul li{position:relative}.pub-list section.year ul li:before{background:rgba(0,0,0,.3);bottom:0;content:" ";left:calc(-2em - 2px);position:absolute;top:-1px;width:2px;z-index:-1}.pub-list section.year ul li:not(.filter-hide):before{top:calc(-1em - 1px)}.pub-list section.year ul li:not(.filter-hide)~li:not(.filter-hide):before{top:-1px}.pub-list section.year:last-child ul li:last-child:before{background:linear-gradient(180deg,rgba(0,0,0,.3) 60%,transparent)}.pub-list .year-mark-wrapper{align-items:center;display:flex;position:sticky;top:5em;width:0}.pub-list .year-mark-wrapper:before{content:"A";visibility:hidden;width:0}.pub-list .year-mark{align-items:center;background:#c5c5c5;border:2px solid #fff;border-radius:10px;display:flex;flex-direction:row;height:10px;justify-content:flex-end;left:-8px;position:relative;width:10px;z-index:2}.pub-list .year-mark:before{content:attr(data-year);font-size:1rem;font-weight:700;margin-right:1em}@media(min-width:62em){.pub-list .year-mark:before{font-size:1.1rem}}.pub-list section.year ul{font-size:1em;margin:0;padding:0 0 0 2em;width:100%}@media(min-width:62em){.pub-list section.year section ul{font-size:1.1em;padding-left:2.2em}}.pub-list section.year ul li{border-bottom:1px dashed #ccc;list-style-type:none;padding:1rem 0 1.25rem}.pub-list section.year ul li:first-child{padding-top:0}.pub-list section.year:last-child ul li:last-child{border-bottom:none}.pub-block{text-align:left}.pub-block .pub-title{font-weight:700}.pub-block .pub-authors,.pub-block .pub-conference,.pub-block .pub-links{font-size:.9em}.pub-block .pub-conference{font-style:italic}.pub-block .pub-conference .pub-conference-acceptance{font-style:normal}.pub-block .pub-abstract-frame{height:0;overflow:hidden;transition:height .4s cubic-bezier(.02,.01,.47,1)}.pub-block .pub-abstract{padding:1em 0;text-align:justify}.pub-block .pub-abstract p:last-child{margin-bottom:0}.pub-block .pub-author.pub-highlight{font-weight:700}.filter-hide{display:none!important}.publist-search-panel .SelectMenu-modal{border-radius:0;width:auto}.publist-search-panel .SelectMenu{min-width:100%}.publist-search-panel details{position:relative}.publist-search-panel .SelectMenu-list--borderless .SelectMenu-divider:first-child{border-top:0}.publist-search-panel .SelectMenu-divider[aria-checked=true] .SelectMenu-icon--check{transform:scale(1);transition:transform .12s cubic-bezier(0,0,.2,1),visibility 0s linear;visibility:visible}.publist-search-panel .SelectMenu-divider{align-items:center;cursor:pointer;display:flex}.publist-search-panel{align-items:flex-start;background:#fff;border-bottom:1px dashed #ccc;display:flex;flex-direction:row;margin-top:-1em;padding:1em 0 .5em;position:sticky;top:0;z-index:11}.publist-search-panel .publist-filters-header{display:flex;flex-direction:column;margin-left:.5em;margin-right:1em}.publist-search-panel .publist-filters-header footer{align-self:center;font-size:small}.publist-filters-header .publist-filters-header-hwrapper{align-items:center;display:flex;height:calc(1.5rem + 4px);justify-content:stretch}.publist-filters-header h4{margin:0;padding:0}.publist-search-panel .publist-filters-row{align-items:center;display:flex;flex:1;flex-wrap:wrap;justify-content:left;margin-bottom:-.5rem;margin-top:-.5rem}.publist-search-panel details{margin:.5rem}.publist-search-panel details>summary[aria-haspopup=true]{align-items:center;cursor:pointer;display:flex;height:1.5rem;outline:none}.publist-search-panel [role=menuitem],.publist-search-panel details summary span{white-space:nowrap}.publist-search-panel .SelectMenu-item .display-value{margin-right:.5rem}.publist-search-panel .SelectMenu-item .Counter{margin-left:auto}.publist-search-panel .dropdown-caret{display:inline-block;height:1.25rem;margin-left:10px;position:relative;width:1.25rem}.publist-search-panel .dropdown-caret:before{left:0;transform:rotate(45deg)}.publist-search-panel .dropdown-caret:after,.publist-search-panel .dropdown-caret:before{background-color:var(--btn-default-border-color);content:"";display:inline-block;height:.1rem;position:absolute;top:.6rem;transition:all .2s ease;width:.75rem}.publist-search-panel .dropdown-caret:after{right:0;transform:rotate(-45deg)}.publist-search-panel details[open] .dropdown-caret:before{transform:rotate(-45deg)}.publist-search-panel details[open] .dropdown-caret:after{transform:rotate(45deg)}
 
@@ -819,6 +819,54 @@ <h4>Filters:</h4>
                         </div>
                     </div>
                 </li>
+                <li data-pub-venue="IEEE TPDS"
+                    data-pub-cat="Journals"
+                    data-pub-extra='{&quot;topic&quot;:[&quot;Disaggregation&quot;],&quot;tag&quot;:[],&quot;badge&quot;:[]}'
+                >
+                    <div class="pub-block">
+                        <div class="pub-title">
+                            <a target="_blank" href="https://doi.org/10.1109/TPDS.2024.3418620">Pyxis: Scheduling mixed tasks in disaggregated datacenters</a>
+                            <span class="Label Label--info">New</span>
+                        </div>
+                        
+                        <div class="pub-authors">
+                            <span class="pub-author pub-author-Sheng-Qi ">Sheng&nbspQi</span>, 
+                            <span class="pub-author pub-author-Chao-Jin ">Chao&nbspJin</span>, 
+                            <span class="pub-author pub-author-Mosharaf-Chowdhury ">Mosharaf&nbspChowdhury</span>, 
+                            <span class="pub-author pub-author-Zhenming-Liu ">Zhenming&nbspLiu</span>, 
+                            <span class="pub-author pub-author-Xuanzhe-Liu ">Xuanzhe&nbspLiu</span>, and 
+                            <span class="pub-author pub-author-Xin-Jin ">Xin&nbspJin</span>
+                        </div>
+                        <div class="pub-conference">
+                            IEEE TPDS 2024, 35(9), 1536-1550
+                            (<a target="_blank" title=""
+                                href="https://ieeexplore.ieee.org/xpl/tocresult.jsp?isnumber=10601540&punumber=71">IEEE TPDS:35(9)</a>)
+                            
+                        </div>
+                        <div class="pub-links">
+                            <a target="_blank" href="https://doi.org/10.1109/TPDS.2024.3418620">[paper]</a>
+                            <a class="pub-link-bibtex" data-clipboard-text="@article{pyxis:tpds,
+    author        = {Sheng Qi and Chao Jin and Mosharaf Chowdhury and Zhenming Liu and Xuanzhe Liu and Xin Jin},
+    title         = {Pyxis: Scheduling Mixed Tasks in Disaggregated Datacenters},
+    journal       = {IEEE Transactions on Parallel and Distributed Systems},
+    year          = {2024},
+    month         = {September},
+    volume        = {35},
+    number        = {9},
+    pages         = {1536-1550},
+}
+">[bibtex]</a>
+                            <a class="pub-link-abstract">[abstract]</a>
+                        </div>
+                        <div class="pub-abstract-frame">
+                            <div class="pub-abstract">
+                                <blockquote><p>Disaggregating compute from storage is an emerging trend in cloud computing. Effectively utilizing resources in both compute and storage pool is the key to high performance. The state-of-the-art scheduler provides optimal scheduling decisions for workloads with homogeneous tasks. However, cloud applications often generate a mix of tasks with diverse compute and IO characteristics, resulting in sub-optimal performance for existing solutions. We present Pyxis, a system that provides optimal scheduling decisions for mixed workloads in disaggregated datacenters with theoretical guarantees. Pyxis is capable of maximizing overall throughput while meeting latency SLOs. Pyxis decouples the scheduling of different tasks. Our insight is that the optimal solution has an “all-or-nothing” structure that can be captured by a single turning point in the spectrum of tasks. Based on task characteristics, the turning point partitions the tasks either all to storage nodes or all to compute nodes (none to storage nodes). We theoretically prove that the optimal solution has such a structure, and design an online algorithm with sub-second convergence. We implement a prototype of Pyxis. Experiments on CloudLab with various synthetic and application workloads show that Pyxis improves the throughput by 3–21× over the state-of-the-art solution.</p>
+
+                                </blockquote>
+                            </div>
+                        </div>
+                    </div>
+                </li>
                 <li data-pub-venue="OSDI"
                     data-pub-cat="Conferences"
                     data-pub-extra='{&quot;topic&quot;:[&quot;Disaggregation&quot;],&quot;tag&quot;:[],&quot;badge&quot;:[]}'
@@ -912,48 +960,46 @@ <h4>Filters:</h4>
                         </div>
                     </div>
                 </li>
-                <li data-pub-venue="IEEE TPDS"
+                <li data-pub-venue="IEEE TASE"
                     data-pub-cat="Journals"
-                    data-pub-extra='{&quot;topic&quot;:[&quot;Disaggregation&quot;],&quot;tag&quot;:[],&quot;badge&quot;:[]}'
+                    data-pub-extra='{&quot;topic&quot;:[&quot;Wide-Area Computing&quot;],&quot;tag&quot;:[],&quot;badge&quot;:[]}'
                 >
                     <div class="pub-block">
                         <div class="pub-title">
-                            <a target="_blank" href="https://doi.org/10.1109/TPDS.2024.3418620">Pyxis: Scheduling mixed tasks in disaggregated datacenters</a>
+                            <a target="_blank" href="https://doi.org/10.1109/TASE.2023.3269639">Fed-ensemble: Ensemble models in federated learning for improved generalization and uncertainty quantification</a>
                             <span class="Label Label--info">New</span>
                         </div>
                         
                         <div class="pub-authors">
-                            <span class="pub-author pub-author-Sheng-Qi ">Sheng&nbspQi</span>, 
-                            <span class="pub-author pub-author-Chao-Jin ">Chao&nbspJin</span>, 
-                            <span class="pub-author pub-author-Mosharaf-Chowdhury ">Mosharaf&nbspChowdhury</span>, 
-                            <span class="pub-author pub-author-Zhenming-Liu ">Zhenming&nbspLiu</span>, 
-                            <span class="pub-author pub-author-Xuanzhe-Liu ">Xuanzhe&nbspLiu</span>, and 
-                            <span class="pub-author pub-author-Xin-Jin ">Xin&nbspJin</span>
+                            <span class="pub-author pub-author-Naichen-Shi ">Naichen&nbspShi</span>, 
+                            <span class="pub-author pub-author-Fan-Lai ">Fan&nbspLai</span>, 
+                            <span class="pub-author pub-author-Raed-Al Kontar ">Raed&nbspAl&nbspKontar</span>, and 
+                            <span class="pub-author pub-author-Mosharaf-Chowdhury ">Mosharaf&nbspChowdhury</span>
                         </div>
                         <div class="pub-conference">
-                            IEEE TPDS 2024, 35(9), 1536-1550
+                            IEEE TASE 2023, 21(3), 2792-2803
                             (<a target="_blank" title=""
-                                href="https://ieeexplore.ieee.org/xpl/tocresult.jsp?isnumber=10601540&punumber=71">IEEE TPDS:35(9)</a>)
+                                href="https://ieeexplore.ieee.org/xpl/RecentIssue.jsp?punumber=8856">IEEE TASE:21(3)</a>)
                             
                         </div>
                         <div class="pub-links">
-                            <a target="_blank" href="https://doi.org/10.1109/TPDS.2024.3418620">[paper]</a>
-                            <a class="pub-link-bibtex" data-clipboard-text="@article{pyxis:tpds,
-    author        = {Sheng Qi and Chao Jin and Mosharaf Chowdhury and Zhenming Liu and Xuanzhe Liu and Xin Jin},
-    title         = {Pyxis: Scheduling Mixed Tasks in Disaggregated Datacenters},
-    journal       = {IEEE Transactions on Parallel and Distributed Systems},
+                            <a target="_blank" href="https://doi.org/10.1109/TASE.2023.3269639">[paper]</a>
+                            <a class="pub-link-bibtex" data-clipboard-text="@article{fed-ensemble:ieee-tase,
+    author        = {Naichen Shi and Fan Lai and Raed Al Kontar and Mosharaf Chowdhury},
+    title         = {Fed-ensemble: Ensemble Models in Federated Learning for Improved Generalization and Uncertainty Quantification},
+    journal       = {IEEE Transactions on Automation Science and Engineering},
     year          = {2024},
-    month         = {June},
-    volume        = {35},
-    number        = {9},
-    pages         = {1536-1550},
+    month         = {July},
+    volume        = {21},
+    number        = {3},
+    pages         = {2792-2803},
 }
 ">[bibtex]</a>
                             <a class="pub-link-abstract">[abstract]</a>
                         </div>
                         <div class="pub-abstract-frame">
                             <div class="pub-abstract">
-                                <blockquote><p>Disaggregating compute from storage is an emerging trend in cloud computing. Effectively utilizing resources in both compute and storage pool is the key to high performance. The state-of-the-art scheduler provides optimal scheduling decisions for workloads with homogeneous tasks. However, cloud applications often generate a mix of tasks with diverse compute and IO characteristics, resulting in sub-optimal performance for existing solutions. We present Pyxis, a system that provides optimal scheduling decisions for mixed workloads in disaggregated datacenters with theoretical guarantees. Pyxis is capable of maximizing overall throughput while meeting latency SLOs. Pyxis decouples the scheduling of different tasks. Our insight is that the optimal solution has an “all-or-nothing” structure that can be captured by a single turning point in the spectrum of tasks. Based on task characteristics, the turning point partitions the tasks either all to storage nodes or all to compute nodes (none to storage nodes). We theoretically prove that the optimal solution has such a structure, and design an online algorithm with sub-second convergence. We implement a prototype of Pyxis. Experiments on CloudLab with various synthetic and application workloads show that Pyxis improves the throughput by 3–21× over the state-of-the-art solution.</p>
+                                <blockquote><p>The increase in the computational power of edge devices has opened up the possibility of processing some of the data at the edge and distributing model learning. This paradigm is often called federated learning (FL), where edge devices exploit their local computational resources to train models collaboratively. Though FL has seen recent success, it is unclear how to characterize uncertainties in FL predictions. In this paper, we propose Fed-ensemble : a simple approach that brings model ensembling to FL. Instead of aggregating local models to update a single global model, Fed-ensemble uses random permutations to update a group of <span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>K</mi></mrow><annotation encoding="application/x-tex">K</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord mathnormal" style="margin-right:0.07153em;">K</span></span></span></span> models and then obtains predictions through model averaging. Fed-ensemble can be readily utilized within established FL methods and does not impose a computational overhead compared with single-model methods. Empirical results show that our model has superior performance over several FL algorithms on a wide range of data sets and excels in heterogeneous settings often encountered in FL applications. Also, by carefully choosing client-dependent weights in the inference stage, Fed-ensemble becomes personalized and yields even better performance. Theoretically, we show that predictions on new data from all <span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>K</mi></mrow><annotation encoding="application/x-tex">K</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord mathnormal" style="margin-right:0.07153em;">K</span></span></span></span> models belong to the same predictive posterior distribution under a neural tangent kernel regime. This result, in turn, sheds light on the generalization advantages of model averaging and justifies the uncertainty quantification capability. We also illustrate that Fed-ensemble has an elegant Bayesian interpretation. Note to Practitioners —provides an algorithm that extracts a set of <span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>K</mi></mrow><annotation encoding="application/x-tex">K</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord mathnormal" style="margin-right:0.07153em;">K</span></span></span></span> solutions without imposing any additional communication overhead in FL. Given multiple solutions, Fed-ensemble can be exploited to personalize inference as well as quantify uncertainty. Such capabilities may be beneficial within multiple practical systems that require uncertainty-aware decision-making. Further, Fed-ensemble may be useful for model validation and hypothesis testing.</p>
 
                                 </blockquote>
                             </div>
@@ -2145,49 +2191,6 @@ <h4>Filters:</h4>
                         </div>
                     </div>
                 </li>
-                <li data-pub-venue="IEEE TASE"
-                    data-pub-cat="Journals"
-                    data-pub-extra='{&quot;topic&quot;:[&quot;Wide-Area Computing&quot;],&quot;tag&quot;:[],&quot;badge&quot;:[]}'
-                >
-                    <div class="pub-block">
-                        <div class="pub-title">
-                            <a target="_blank" href="https://doi.org/10.1109/TASE.2023.3269639">Fed-ensemble: Ensemble models in federated learning for improved generalization and uncertainty quantification</a>
-                            
-                        </div>
-                        
-                        <div class="pub-authors">
-                            <span class="pub-author pub-author-Naichen-Shi ">Naichen&nbspShi</span>, 
-                            <span class="pub-author pub-author-Fan-Lai ">Fan&nbspLai</span>, 
-                            <span class="pub-author pub-author-Raed-Al Kontar ">Raed&nbspAl&nbspKontar</span>, and 
-                            <span class="pub-author pub-author-Mosharaf-Chowdhury ">Mosharaf&nbspChowdhury</span>
-                        </div>
-                        <div class="pub-conference">
-                            IEEE TASE 2023
-                            (<a target="_blank" title=""
-                                href="https://ieeexplore.ieee.org/xpl/RecentIssue.jsp?punumber=8856">IEEE TASE</a>)
-                            
-                        </div>
-                        <div class="pub-links">
-                            <a target="_blank" href="https://doi.org/10.1109/TASE.2023.3269639">[paper]</a>
-                            <a class="pub-link-bibtex" data-clipboard-text="@article{fed-ensemble:ieee-tase,
-    author        = {Naichen Shi and Fan Lai and Raed Al Kontar and Mosharaf Chowdhury},
-    title         = {Fed-ensemble: Ensemble Models in Federated Learning for Improved Generalization and Uncertainty Quantification},
-    journal       = {IEEE Transactions on Automation Science and Engineering},
-    year          = {2023},
-    month         = {May},
-}
-">[bibtex]</a>
-                            <a class="pub-link-abstract">[abstract]</a>
-                        </div>
-                        <div class="pub-abstract-frame">
-                            <div class="pub-abstract">
-                                <blockquote><p>The increase in the computational power of edge devices has opened up the possibility of processing some of the data at the edge and distributing model learning. This paradigm is often called federated learning (FL), where edge devices exploit their local computational resources to train models collaboratively. Though FL has seen recent success, it is unclear how to characterize uncertainties in FL predictions. In this paper, we propose Fed-ensemble : a simple approach that brings model ensembling to FL. Instead of aggregating local models to update a single global model, Fed-ensemble uses random permutations to update a group of <span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>K</mi></mrow><annotation encoding="application/x-tex">K</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord mathnormal" style="margin-right:0.07153em;">K</span></span></span></span> models and then obtains predictions through model averaging. Fed-ensemble can be readily utilized within established FL methods and does not impose a computational overhead compared with single-model methods. Empirical results show that our model has superior performance over several FL algorithms on a wide range of data sets and excels in heterogeneous settings often encountered in FL applications. Also, by carefully choosing client-dependent weights in the inference stage, Fed-ensemble becomes personalized and yields even better performance. Theoretically, we show that predictions on new data from all <span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>K</mi></mrow><annotation encoding="application/x-tex">K</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord mathnormal" style="margin-right:0.07153em;">K</span></span></span></span> models belong to the same predictive posterior distribution under a neural tangent kernel regime. This result, in turn, sheds light on the generalization advantages of model averaging and justifies the uncertainty quantification capability. We also illustrate that Fed-ensemble has an elegant Bayesian interpretation. Note to Practitioners —provides an algorithm that extracts a set of <span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>K</mi></mrow><annotation encoding="application/x-tex">K</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.68333em;vertical-align:0em;"></span><span class="mord mathnormal" style="margin-right:0.07153em;">K</span></span></span></span> solutions without imposing any additional communication overhead in FL. Given multiple solutions, Fed-ensemble can be exploited to personalize inference as well as quantify uncertainty. Such capabilities may be beneficial within multiple practical systems that require uncertainty-aware decision-making. Further, Fed-ensemble may be useful for model validation and hypothesis testing.</p>
-
-                                </blockquote>
-                            </div>
-                        </div>
-                    </div>
-                </li>
                 <li data-pub-venue="NSDI"
                     data-pub-cat="Conferences"
                     data-pub-extra='{&quot;topic&quot;:[&quot;Energy-Efficient Systems&quot;,&quot;Systems + AI&quot;],&quot;tag&quot;:[],&quot;badge&quot;:[]}'
@@ -5605,7 +5608,7 @@ <h4>Filters:</h4>
         </section>
     </div> <!-- pub-list -->
 </div>
-<!-- end-publist-4c41a19c -->
+<!-- end-publist-c442244d -->
 
 <script data-pjax src="/assets/publist/main.js"></script>