Skip to content

show the Boreas tar process sizing and storage-first worker order in the deck #60

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
zaihuaji merged 1 commit into from
Oct 19, 2026
Merged

show the Boreas tar process sizing and storage-first worker order in the deck #60

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Binary file modified docs/slides/dsquasar_guide.pdf
View file
Open in desktop
Binary file not shown.
Binary file modified docs/slides/dsquasar_guide.pptx
View file
Open in desktop
Binary file not shown.
22 changes: 11 additions & 11 deletions docs/slides/gen_dsquasar.js
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1161,10 +1161,10 @@ function reftable(s, x, y, w, colW, rows, hdr, hcolor, kcolor, fs) {
fontFace:SANS, bold:true, fontSize:15, color:DEEP, margin:0 });
s.addText("One child per tar file \u2014 many small, cheap units.", {
x:0.75, y:2.85, w:5.6, h:0.3, fontFace:SANS, fontSize:11.5, color:MUTE, margin:0 });
code(s, 0.75, 3.2, 5.6, 0.85,
"MPTLIMIT = 200 Glade tar files per process\nMPOLIMIT = 20 object store tar files per process\nMPTMAX = 12 process ceiling", 11.5);
s.addText("A small divisor with a high cap: lots of tar work should spread wide, object store tars (download first) widest.", {
x:0.75, y:4.12, w:5.6, h:0.6, fontFace:SANS, fontSize:11, color:INK,
code(s, 0.75, 3.18, 5.6, 1.08,
"MPTLIMIT = 200 Glade tar files per process\nMPOLIMIT = 20 Boreas tar files per process\nMPTMAX = 12 process ceiling\nprocs = ceil(Glade/200 + Boreas/20)", 11);
s.addText("A small divisor with a high cap: tar work spreads wide, Boreas tars ten times wider.", {
x:0.75, y:4.34, w:5.6, h:0.4, fontFace:SANS, fontSize:11, color:INK,
margin:0, valign:"top" });

// uploading card
Expand All @@ -1180,9 +1180,9 @@ function reftable(s, x, y, w, colW, rows, hdr, hcolor, kcolor, fs) {
x:7.15, y:4.05, w:5.5, h:0.6, fontFace:SANS, fontSize:11, color:INK,
margin:0, valign:"top" });

s.addText("Why the two sets of numbers", { x:0.5, y:5.0, w:12.35, h:0.35,
s.addText("Why Boreas tars get more processes", { x:0.5, y:5.0, w:12.35, h:0.35,
fontFace:SANS, bold:true, fontSize:15, color:INK, margin:0 });
s.addText("Both phases once shared a single limit. Tarring came out under-provisioned while uploading reserved cpus it could never use \u2014 cpus that PBS charges for and that other jobs then cannot have. Splitting the constants fixed both ends.", {
s.addText("Every member of a Boreas tar is staged down from the object store and checksummed before it is tarred, so it takes many times longer than a Glade tar. The backlog sat in one dataset, d694517 (~730 GB, ~146 tars): one process per 200 tars gave it 1, per 20 it gets 8.", {
x:0.5, y:5.35, w:12.35, h:0.55, fontFace:SANS, fontSize:12, color:MUTE,
margin:0, valign:"top", lineSpacingMultiple:1.05 });
s.addShape(p.ShapeType.roundRect, { x:0.5, y:6.0, w:12.35, h:0.6,
Expand Down Expand Up @@ -1280,17 +1280,17 @@ function reftable(s, x, y, w, colW, rows, hdr, hcolor, kcolor, fs) {
margin:0, valign:"top", lineSpacingMultiple:1.05 });
reftable(s, 6.9, 3.42, 5.95, [1.35, 4.6], [
["slot 1", "plain argv \u2014 repeat cron still blocks as a duplicate"],
["odd", "walks datasets in ascending dsid order"],
["even", "walks datasets descending, from d999999"],
["odd", "Boreas datasets first, then Glade; dsid ascending"],
["even", "Glade datasets first, then Boreas; dsid descending"],
], ["Slot", "Behaviour"], DEEP);
s.addText("A pair of workers therefore meets in the middle rather than fighting over the same dataset lock from the same end of the list.", {
s.addText("A pair of workers starts on different storage and meets in the middle rather than fighting over the same dataset lock.", {
x:6.9, y:5.15, w:5.95, h:0.5, fontFace:SANS, fontSize:11.5, color:INK,
margin:0, valign:"top", lineSpacingMultiple:1.05 });
s.addShape(p.ShapeType.roundRect, { x:6.9, y:5.72, w:5.95, h:0.85,
rectRadius:0.08, fill:{color:MID}, line:{type:"none"} });
s.addText([
{ text:"Not a fix for one huge dataset. ", options:{ bold:true, color:AMBERLT } },
{ text:"Locks are per dataset, so a second worker cannot help inside the dataset that is actually stuck.", options:{ color:"C3D7EE" } },
{ text:"One huge dataset needs more processes. ", options:{ bold:true, color:AMBERLT } },
{ text:"Locks are per dataset, so a second worker cannot help inside it; more tar processes (MPOLIMIT) do.", options:{ color:"C3D7EE" } },
], { x:7.1, y:5.72, w:5.55, h:0.85, fontFace:SANS, fontSize:11,
margin:0, valign:"middle", lineSpacingMultiple:1.05 });
foot(s);
Expand Down
Loading