Creating this page to more easily track and visualize what system/group/wave is being used where in the ~ipp/ippconfig/pantasks_hosts.input file. seem to recall was done before but cannot find ...

Node group summary:

node group	hostnames	typical cores	typical RAM	comments
wave1 (w1)	ipp008--021	8 cores	24G
wave2 (w2)	ipp023--036	8 cores	32G
wave3 (w3)	ipp037--053	8 cores	24/32G
wave4 (w4)	ipp054--066	12 cores	49G
compute (c)	ippc01--c19	8 cores	32G	many are ignored_
compute2 (c2)	ippc20--c29	12 cores	24G (c26-c28 49G, c29 73G)
compute3 (c3)	ippc30--c63	12 cores	49G
stare (s)	stare00--04	12 cores	97G
stsci (st)	stsci00--19	12 cores	49G	not to be used in processing

Other special (sub)groups:

• dvo/ignore: ipp004-007
• apache: ippc01-c09

• wave1_weak -- weak meaning cannot have full loads (i.e., lost CPU or other issue): ipp010-012,018
• wave2_weak: ipp023,027,028,029,032,033
• wave3_weak: ipp047
• compute3_weak:

• compute2_himem -- c2 w/ extra RAM: ippc26-c28, c29 (c29 typically out of processing for very deep stack tests)

• deepstack: typically used by deepstack pantasks w/ 2x c3, but modified again for other uses (LAP staticsky 3x c3, 4x stare)

2/13/2014 nightly+LAP+staticsky

Gene heavily shuffled allocation, heavily weighted to staticsky and stack. Few weak systems being fully loaded, crash+burn and taking out processing. Since only few, taking back out of processing (MEH 2/19)

2/4/2014 nightly+LAP+staticsky

Want to balance for more staticsky progress now.

• ipptopsps also will not be running for a bit so can use more of the system.
• have noticed staticsky reaching >10G, will want to keep off lower RAM systems with data.
• c was overloaded before, ease up some on it since it hosts apache as well, c2 also could be overloaded some with stacks as was in past
• suspect w1-4 could be loaded further but should monitor nightly processing use first
• want to wait for full cab/serial control mapping before overloading key nodes, if go down want to be able to reboot them quickly

pantasks	w1	w1w	w2	w2w	w3	w3w	w4	c	c2	c3	c3w	stare	comment
stdscience	5	2	4	2	4	3	4	4	2	2	3		100% use, LAP + updates light, nightly moder, N=291? (need >200)
stack					1		2	1	3				100% use, LAP/nightly heavy, N~73 (need >60)
summitcopy			1	1	1		1						nightly, need N>30
registration	1		1	1	1		1						nightly
distribution			1	1	1			1					nightly
pstamp	2							2					regular
cleanup	1		1	1	1		1						regular
publishing								1					nightly, light
ippsky/staticsky										3		4	100% use, LAP, heavy+very long running, N=127 (+20*80% for overloaded stare) ?
equiv. totals	x	x	x	x	x	x	x	x	x	x	x	x
replication			1	1	1	1							light load, not always in use -- rebalance when in use
deepstack										2			heavy+very long running, not always in use, -2x c3 from stack/staticsky
detrend	1		1		1		1						light load, not always in use -- rebalance when in use

Load level

• light = ~1 thread equiv, fairly quick runtime (x1)
• moder = ~1-4 thread, moderately fast runtime (x1-2)
• heavy = mostly 2-4 thread, longer runtime (x3-4)

When ipptopsps loaders not in use -- ipp045-ipp066 (half wave3+wave4), stare

• overloaded stare (+4x) nodes for staticsky

Cleanup possible to add to stsci nodes for local delete?

---

late 2013

A reconfiguration once ipptopsps and loaders running full time.

• stare fully allocated to PSPS
• w1-4 reduced to account for extra load to data nodes

pantasks	w1	w1w	w2	w2w	w3	w3w	w4	c	c2	c3	c3w	stare	comment
stdscience	3	2	3	2	3	3	3	4	4	5	3
stack					2		2	2	2	2
summitcopy			1	1	1		1
registration	1		1	1	1		1
distribution			1	1	1			1
pstamp	2							2
cleanup	1		1	1	1
publishing								1
update								1	1
equiv. totals	x	x	x	x	x	x	x	x	x	x	x	x
replication			1	1	1	1
deepstack										2
detrend	1		1		1		1

---

---