Robotics and Tape Drives

1
Robotics and Tape Drives

• Activities
• (and non-activities)
• 31 January 2006

2
Robotics

• We have 2 sites presently, 513 (large) and 613
  (small)
• There are 3 library types IBM 3584, STK SL8500,
  STK Powderhorn
• The largest is the oldest, the 5 plus 5
  Powderhorns (60,000 slots)
• The newest are the fully configured IBM 3584,
  and the STK SL8500
• These were used (to some extent) in 4Q05, and
  this will continue in 2006
• IBM library is 6000 slots, 3 PB max, with IBM
  3592 E05 (40)
• STK SL8500 is 7500 slots, will be 3.5 PB max
  when T10000s arrive (40)
• The Powderhorn libraries are large but only have
  a future with the T10000.
• They do not support the IBM 3592 E05 or LTO
  drives.
• They would represent a useful 30 PB, but
  end-of-life around 2010..

3
IBM 3584
4
STK SL8500
5
513 Powderhorns, IBM 3584
6
613 Powderhorns
7
Tight(ish) squeeze.
8
Drives
We have 5 drive types presently IBM 3592 E05,
500 GB, 100 MB/s, 40 units here but only 250
cartridges. STK T10000, 2 fat test drives, 500
GB, 120 MB/s, and 1500 cartridges here. We
expect 40 production T10000. These IBM and STK
products are the newest. Both have been tested
3-4Q 05. Tests (and limited production) should
resume .. when tape servers are installed ..
STK drives arrive We also have 8 IBM 3592, 300
GB, 40 MB/s in a small 3584 (to TSM..) We also
have 6 HP LTO3, 300 GB, 80 MB/s in the SL8500
system. ??. The oldest? STK 9940B, 200 GB and
30 MB/s, are still the main production drive
(44) Retirement starts 3Q06?
9
Vexations
We have the wrong network from disk server to
tape server (Gbit ethernet) We have the wrong FC
adapters (new drives will be 4 GB) We have the
wrong disk servers (single 100 MB/s streams not
easy) Do we still have a sensible architecture
(disk server layer, tape server layer)? Next
drives will be even faster, more embarrassing,
and they are not that far off.. LTO-4 in
1-2Q07 Upgrades of top line drives happens every
2 years or so, 2007-8 for IBM and STK?
10
Is it confusion, or purpose?
Confusion? Complete IBM library, 40 drives, but
very little media Complete SL8500 library
(almost..), planned amount of media, but only
test drives A lot of unused Powderhorn
equipment A shortage of tape servers (hopefully
soon to be remedied, but) Purpose? The
Powderhorn will support T10000 Both 3584 and
SL8500 can support LTO n It was hoped to test
new tape layer architectures, FC fabrics, ..
11
And, as we move forward .
Need to use new media, drives, libraries, tape
servers for new data ASAP We have 3,000 media,
or 1.5 PB, and existing CASTOR is 4.5 PB. Need
to use new drives (both) for forthcoming Data
Challenges Prepare for call for tenders for more
equipment before end 2006 Install this equipment
end 2006 or early 2007 by luck, have a very
wide range of options NO candidate shown yet
to be overwhelmingly good, or to have definitely
failed STILL have LTO n option (3584 or
SL8500) Powderhorns can carry on for longer, but
need the T10000
12
Perhaps more difficult.
Need to improve rmcdaemon/smc IBM robot
control Ignore nonsense replies (access
geometry failure,), add remote move media,
remove incorrect drive DOWN Need to improve
efficiency of drive use in CASTOR (or spend
more) As low as 5 effective for READ, poor for
WRITE This is software, where there is a
distinct lack of personnel (and time) Select
drive and media for next read/write, and
pre-position media, 60s saved Writing tapes,
fill to near BOT (all drives), 90s saved Write
all small files in BOT turn-around area, drives
look almost like 9840 Write small files on
fastest label writing drive (IBM?), 6s saved per
file Read files in order from BOT, NOT sequence
order, reads are 2-3 times as fast