Easy Tier: automatic data placement

Easy Tier needs days of real I/O before you see its effect, so no laptop lab can show it migrating. What you’ll do instead is read Easy Tier’s settings, see why a pool needs more than one rank, and use a small Python model to see why moving heat is worth it. Start in C:\ds8k-lab (shell setup is in Lesson 1).

Practice shell note: the practice shell is our own teaching tool, not IBM software, and real DS CLI output looks very close to this without being identical. The shell shows Easy Tier settings but doesn’t run Easy Tier or the chsi and tier commands, so you’ll only see those as real syntax here. CMUN02037E in the break lab is a practice shell message.

1Read the Easy Tier settings

Type these at the dscli> prompt:

simctl reset
showsi

You should see this:

dscli> showsi
Date/Time: 24 October 2026 13:55:48 UTC IBM DSCLI Version: 7.10.10.348 DS: IBM.2107-78SV001
Name             ds8k-g10-lab
...
Model            A05
...
ETAutoMode       tiered
ETMonitor        automode
IOPMmode         Disabled
ETCCMode         -
ETHMTMode        Enabled
ETSRMode         Enabled
ETTierOrder      High performance
ETAutoModeAccel  Disabled
...

The shortened showsi output is where Easy Tier lives at system level. ETAutoMode tiered means automatic mode only manages multitier pools, and ETMonitor automode means Easy Tier monitors the volumes in the pools it manages.

ETSRMode Enabled reserves space for extent moves, ETTierOrder High performance sends new data to the fastest tier first (the all-flash default), and ETHMTMode Enabled allows heat maps to go to a remote DS8000.

ETAutoModeAccel Disabled is the normal state, since acceleration is only a short-term boost. IOPMmode Disabled refers to the old I/O Priority Manager, which has been discontinued since Release 9.

✅ Checkpoint: You can say what ETAutoMode tiered means.

2See how you would change them on a real DS8000

The practice shell doesn’t include chsi, so only read these. It’s the verified DS CLI syntax for a real system:

chsi -etautomode all -etmonitor all IBM.2107-78SV001
chsi -ettierorder highutil IBM.2107-78SV001
manageextpool -action etmigpause -duration 1H IBM.2107-78SV001/P0
managefbvol -action tierassign -tier flashtier0 2000
offloadfile -etdataCSV C:\temp

Line 1 lets Easy Tier manage and monitor every pool, single-tier ones included, so it can auto-rebalance them. Line 2 changes where new data lands, line 3 pauses migrations in pool P0 for an hour, line 4 pins volume 2000 to the top tier, and line 5 downloads the Easy Tier report CSV files and Excel tool.

On a real system chsi replies with a CMUC00042I “successfully modified” message.

✅ Checkpoint: You can name the command that changes Easy Tier settings for the whole system.

3One rank: nothing to balance

Save this as C:\ds8k-lab\setup-16.txt, then quit, run it with -script like in earlier lessons and log in again. It builds two Flash Tier 2 ranks but only puts R0 in pool P0.

mkarray -raidtype 6 -arsite S7
mkarray -raidtype 6 -arsite S8
mkrank -array A0 -keygrp 1 -stgtype fb
mkrank -array A1 -keygrp 1 -stgtype fb
mkextpool -rankgrp 0 -keygrp 1 -stgtype fb t2_0
chrank -extpool P0 R0

Type these at the dscli> prompt:

mkfbvol -extpool P0 -cap 100 -name db_#h 2000
showfbvol -rank 2000

You should see this:

dscli> mkfbvol -extpool P0 -cap 100 -name db_#h 2000
CMUC00025I mkfbvol: FB volume 2000 successfully created.
dscli> showfbvol -rank 2000
Date/Time: 24 October 2026 13:56:33 UTC IBM DSCLI Version: 7.10.10.348 DS: IBM.2107-78SV001
...
eam                        rotateexts
...
==============Rank extents==============
rank extents capacity (MiB/cyl) metadata
========================================
R0   6400    102400             yes

All 6,400 extents of the volume are on R0, because R0 is the only rank, so if this volume gets busy R0 gets busy and Easy Tier has nowhere to move anything. That’s why the Redbook tells you to put more than one rank in a pool if you want Easy Tier to optimize rank use.

✅ Checkpoint: All 6,400 extents are on R0.

4Add a second rank

Type these at the dscli> prompt:

chrank -extpool P0 R1
mkfbvol -extpool P0 -cap 100 -name db_#h 2001
showfbvol -rank 2001
showextpool P0

You should see this:

dscli> chrank -extpool P0 R1
CMUC00008I chrank: Rank R1 successfully modified.
dscli> mkfbvol -extpool P0 -cap 100 -name db_#h 2001
CMUC00025I mkfbvol: FB volume 2001 successfully created.
dscli> showfbvol -rank 2001
Date/Time: 24 October 2026 13:56:34 UTC IBM DSCLI Version: 7.10.10.348 DS: IBM.2107-78SV001
...
eam                        rotateexts
...
==============Rank extents==============
rank extents capacity (MiB/cyl) metadata
========================================
R0   3200    51200              yes
R1   3200    51200              no
dscli> showextpool P0
Date/Time: 24 October 2026 13:56:34 UTC IBM DSCLI Version: 7.10.10.348 DS: IBM.2107-78SV001
Name                      t2_0
...
numranks                  2
...
numtiers                  1
etmanaged                 yes
extsize                   16MiB
...

The new volume is split 3,200 and 3,200 across R0 and R1, while volume 2000 is still only on R0, since the practice shell doesn’t run Easy Tier and nothing moves it. On a real DS8000 with Easy Tier managing this pool, auto-rebalance would shift some of 2000’s hot extents to R1 over time.

showextpool (shortened) confirms numtiers 1, meaning a single-tier pool, and etmanaged yes.

✅ Checkpoint: volume 2001 has 3,200 extents on each rank.

5See why moving heat pays off

Real workloads are skewed, with a few extents doing most of the work, and this small model (ours, not IBM’s algorithm) shows what that means. Save it as C:\ds8k-lab\heat.py, next to ds8sim.py.

It runs right there on your laptop, not on the DS8000. The model part doesn’t need a DS8000 at all, and the --check part reads your pools through the DS CLI the way outside automation would.

The script:

# heat.py - why Easy Tier works: a toy model (ours, not IBM's algorithm) plus a pool check
# Runs on your workstation or jump server, not on the DS8000.
# The model needs no DS8000. --check talks to it only through the DS CLI (auto.profile: banner off, header on, format delim).
import csv, random, subprocess, sys
DSCLI = ["python", "ds8sim.py", "-cfg", "auto.profile"]   # at work: ["dscli", "-cfg", "auto.profile"]

def dscli(*args):
    """Run one DS CLI command and return its rows as dictionaries."""
    out = subprocess.run(DSCLI + list(args), capture_output=True, text=True).stdout
    lines = [l for l in out.splitlines() if l and not l.startswith("SaatVedha") and not l.startswith("CMUC")]
    return list(csv.DictReader(lines))

def model():
    # 1,000 extents; a few are very busy ("hot"), most are quiet. Skew is normal in real workloads.
    random.seed(16)
    extents = 1000
    fast_slots = 100                       # the fast tier holds 10% of the extents
    # I/O count per extent in one day: a small hot set, a warm middle, a long cold tail
    io = [random.randint(5000, 9000) if i < 50 else
          random.randint(500, 1500) if i < 200 else
          random.randint(0, 60) for i in range(extents)]
    random.shuffle(io)                     # data lands in random places at first
    total = sum(io)
    def fast_share(placement):
        return 100 * sum(io[i] for i in placement) / total
    first_100 = range(fast_slots)                                   # whatever landed there first
    hottest_100 = sorted(range(extents), key=lambda i: io[i], reverse=True)[:fast_slots]
    print(f"Total I/Os per day         : {total:,}")
    print(f"Fast tier before learning  : {fast_share(first_100):5.1f}% of I/O")
    print(f"Fast tier after moving heat: {fast_share(hottest_100):5.1f}% of I/O")
    top5 = sorted(io, reverse=True)[:extents // 20]
    print(f"Busiest 5% of extents carry: {100 * sum(top5) / total:5.1f}% of I/O")

def check():
    # Can Easy Tier move heat in each pool? It needs more than one rank (or more than one tier).
    ranks = dscli("lsrank", "-l")
    for p in dscli("lsextpool"):
        pool = dscli("showextpool", p["ID"])[0]
        mine = [r for r in ranks if r["extpoolID"] == p["ID"]]
        print(f"Pool {p['ID']} {p['Name']}: {pool['numranks']} rank(s), {pool['numtiers']} tier(s), "
              f"Easy Tier managed {pool['etmanaged']}")
        for r in mine:
            print(f"  {r['ID']} used extents {int(r['usedexts']):>6,}")
        if int(pool["numranks"]) < 2 and int(pool["numtiers"]) < 2:
            print("  -> one rank, one tier: Easy Tier has nowhere to move hot extents")
        else:
            print("  -> Easy Tier can move hot extents between ranks or tiers here")

if "--check" in sys.argv:
    check()
else:
    model()

For the check you need a saved password and a profile a script can read. Use the password you picked in Lesson 1, and if the shell says the record already exists, you’ve saved it before, so carry on.

Type these at the dscli> prompt:

managepwfile -action add -name admin -pw <your password> -pwfile security.dat
quit

You should see this:

dscli> managepwfile -action add -name admin -pw <your password> -pwfile security.dat
CMUC00206I managepwfile: Record 10.0.0.1/admin successfully added to password file security.dat.
dscli> quit

Save this as auto.profile in the lab folder:

hmc1: 10.0.0.1
username: admin
pwfile: security.dat
devid: IBM.2107-78SV001
banner: off
header: on
format: delim
delim: ,

Banner off, header on and format delim make the DS CLI print one header row followed by plain comma-separated rows, which is easy for a script to read. Implementation Lesson 2 explains every line.

Run this:

python heat.py
python heat.py --check

You should see this:

Total I/Os per day         : 512,602
Fast tier before learning  :   7.3% of I/O
Fast tier after moving heat:  80.0% of I/O
Busiest 5% of extents carry:  67.1% of I/O
Pool P0 t2_0: 2 rank(s), 1 tier(s), Easy Tier managed yes
  R0 used extents  9,600
  R1 used extents  3,200
  -> Easy Tier can move hot extents between ranks or tiers here

With random placement the fast tier (10% of capacity) serves only about 7% of the I/O, but once the hottest 100 extents move there, the same 10% serves 80%. That’s what Easy Tier is about: a small amount of the fastest flash can carry most of the work as long as the right extents live on it.

The check ran lsextpool, showextpool P0 and lsrank -l for you. P0 has two ranks, so Easy Tier has somewhere to move heat. R0 holds 9,600 used extents (all of volume 2000 plus half of 2001) and R1 holds 3,200, and on a real DS8000 auto-rebalance would even that out by heat over time.

Change random.seed(16) to some other number and run it again. The exact values will change but the pattern stays the same. At work a scheduler on the jump server can run python heat.py --check every morning and mail the report, so a pool left with a single rank gets noticed.

✅ Checkpoint: Your script shows 80.0% after moving heat, and the check shows 2 ranks in P0.