Chapter 2. FICON 13
Figure 2-9 Impact of FICON on highly parallel table scans, using two ESSs
The two ESSs were independent of each other. They were using different channels and
ESCON directors. Therefore, the throughput with two ESSs and 32 ESCON channels was
double that of one ESS and 16 ESCON (412 versus the 206 MB/sec in Figure 2-8 on
page 12). In all of these ESCON cases, the channel utilizations were close to 99%, indicating
that additional storage servers could not provide higher throughput if they shared the same
channels.
As was the case with one ESS, a four to one reduction in the number of channels with two
ESSs results in only a 10% reduction in the throughput (372 versus 412 MB/sec), again with
twice as many FICON HAs as FICON channels. Furthermore, the increase in the throughput
gained by adding a second ESS sharing the same set of 8 FICON channels increased by
52% (372 versus 245 MB/sec in Figure 2-8 on page 12), compared to only the 5% increase
observed with 16 ESCON channels (217 versus 206 MB/sec in Figure 2-8 on page 12). This
proves that the 8 FICON channels were not saturated when serving one ESS, but the 16
ESCON channels were nearly so.
2.3.4 DB2 utilities
For the DB2 utilities, the ESCON measurements were done with all 16 ESCON channels
online, and the FICON measurements were done with all 8 FICON channels online. However,
neither the ESCON nor FICON channels were constrained.
COPY utility
The COPY utility was first measured with FICON channels without striping. Then two DFSMS
stripes were defined for the image copy output data set in order to demonstrate how striping
can be used to further improve performance.
Figure 2-10 shows that with 8 FICON channels, the COPY utility provides 34% more
throughput than with 16 ESCON channels.
An additional 36% improvement can be obtained with striping.
217
412
204
372
16 ESCON 32 ESCON 4 FICON (D) 8 FICON (F)
0
100
200
300
400
500
MB/sec
Scan of 60 partitions - 2 ESS
Throughput (MB/sec)
14 DB2 for z/OS and OS/390 Version 7 Selected Performance Topics
Figure 2-10 Impact of FICON on the COPY utility
RECOVER utility
Figure 2-11 shows that with 8 FICON channels, the RECOVER utility provides 56% more
throughput than with 16 ESCON channels.
Figure 2-11 Impact of FICON on the RECOVER utility
COPY utility
Throughput (MB/sec)
14
18.8
14.3
25.6
NO STRIPING TWO STRIPES
0
5
10
15
20
25
30
MB/sec
16 ESCON
8 FICON
34% improvement
with no striping
79% improvement
with 2 stripes
RECOVER utility
Throughput (MB/sec)
9
14
16 ESCON 8 FICON
0
5
10
15
MB/sec
56% improvement
Chapter 2. FICON 15
LOAD utility
The LOAD utility was measured using a 1.2 GB table (10 million rows, 117 byte rows, and 26
columns per row). One index containing two columns was defined. Figure 2-12 shows that
each of the LOAD utility phases improves the elapsed time and, overall, the LOAD utility runs
31% faster.
Figure 2-12 Impact of FICON on the LOAD utility
The CPU time for ESCON and FICON is the same (111 sec). The I/O time is still higher than
the CPU time on zSeries 900, but the ratio of CPU time to elapsed time is reduced with
FICON. With many partitions, it is likely that the LOAD utility will become CPU bound.
REORG utility
The elapsed time of the REORG utility (with SORTDATA specified) was measured using the
same table as described for the LOAD utility, with a 50% cluster ratio.
REORG was also measured with two DFSMS stripes for the SYSREC data set, to
demonstrate how striping can be used to further reduce the elapsed time.
Again, FICON improved all phases of REORG elapsed time. Figure 2-13 shows that the
REORG utility runs 35% faster with FICON. An additional 14% reduction can be obtained with
striping.
LOAD utility with index
Elapsed time (sec)
16 ESCON 8 FICON
0
50
100
150
200
250
sec
Build Phase
Sort Phase
Load Phase
31% faster
217
150

Get DB2 for z/OS and OS/390 Version 7 Selected Performance Topics now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.