Académique Documents
Professionnel Documents
Culture Documents
There are two categories for the R7.4 performance enhancements; 32 thread and
64 thread.
Distribute the work to increase parallelism and limit the effect of synchronization to shared data
structures.
Algorithms to cache updates to LRU lists in CPU caches without affecting LRU algorithm.
Multi-threading background processes to keep with foreground I/O.
Splitting hot locks into multiple locks to reduce lock contention.
Enhancements allow for both overall increased system IOps performance as well
as reduced response times at higher IOps rates.
2014 IBM Corporation
R7.4 Performance
All results are from a DS8870 Model 961 High Performance All Flash
Configuration with 16 P7+ cpu cores.
All results are on pre-GA code and are subject to change.
Performance Benchmarks
Sequential IO
Large block sequential reads or writes to the storage servers drives. Similar to SPC-2.
Corner benchmarks
Simulates IO done by OLTP applications. Similar to SPC-1. Do a mix of random reads and writes
with cache hits and cache misses.
DBO(Database Open): 70% read/30% write, 4KB IOs, 50% read cache hit
DB zOS(Database System Z): 75% read/25% write, 4KB IOs, 72% read cache hit
Cache Hostile: 72% read/28% write, 4KB IOs, 40% read cache hit
Perform just one type of IO pattern. Most applications combine a mix of different IO patterns.
Cache Hits: 4KB random reads or writes to the storage servers cache.
Cache Read Misses: 4KB random reads to the storage servers drives.
Writes Miss: 4KB random writes to the storage servers drives.
Sequential
800
700
600
500
400
300
200
100
0
25
20
GB/s
KIO/s
4KB Random
10
5
0
Read
R7.3
15
R7.4 32 thread
Read
Write
R7.4 64 thread
R7.3
R7.4 32 thread
Write
R7.4 64 thread
DBO(70/30/50) w/ 8 HPFE
R7.3 vs R7.4 w/ 32 and 64 threads
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0
100
200
300
400
500
600
700
800
900
1000
K IOps
R7.3
R7.4 32 threads
R7.4 64 threads
The DBO benchmark does 70% reads/30% writes with 50% read cache hits.
DBO(70/30/50)
R7.3 vs R7.4 w/ 64 threads for 1 to 8 HPFE
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0
100
200
300
400
500
600
700
800
900
1000
K IOps
1 HPFE R7.3
2 HPFEs R7.3
4 HPFEs R7.3
8 HPFEs R7.3
1 HPFE R7.4
2 HPFE R7.4
4 HPFE R7.4
8 HPFE R7.4
The DBO benchmark does 70% reads/30% writes with 50% read cache hits.
0.6
0.4
0.2
0
0
100
200
300
400
500
600
700
800
K IOps
R7.4 32 threads
R7.4 64 threads
The DB z/OS benchmark does 75% reads/25% writes with 72% read cache hits.
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0
100
200
300
400
500
600
700
800
K IOps
R7.4 32 threads
R7.4 64 threads
The Cache Hostile benchmark does 72% reads/28% writes with 40% read cache hits.
KIO/s
4KB Random
700
600
500
400
300
200
100
0
Read
R7.4 32 thread
10
R7.4 64 thread
2014 IBM Corporation
Multi-Target
PPRC
11
Multi-Target PPRC
Configuration
12
13
14
15
16
Online
Batch
Read hit
92%
92%
Read/Write Ratio
3:1
2.4:1
% Sequential Read
23
Destage Rate
8.4%
16.5%
IOPS
Curve
45KIOPs/80KIOPs
50KB
40KB
560
27KB
480
17
18
19
New Drives
20
New Drives
21
22
23
24
25
Disk Magic
Support
26
Delivery
27
Final Release to be delivered by Jan 9th, will add Multi-target PPRC and
1.6TB 2.5 SSDs
All statements regarding IBM future direction and intent are subject to change or withdrawal without notice, and represent goals and objectives only.
Some information addresses anticipated future capabilities. Such information is not intended as a definitive statement of a commitment to specific levels of performance,
function or delivery schedules with respect to any future products. Such commitments are only made in IBM product announcements. The information is presented here
to communicate IBM's current investment and development activities as a good faith effort to help with our customers' future planning.
Performance is based on measurements and projections using standard IBM benchmarks in a controlled environment. The actual throughput or performance that any
user will experience will vary depending upon considerations such as the amount of multiprogramming in the user's job stream, the I/O configuration, the storage
configuration, and the workload processed. Therefore, no assurance can be given that an individual user will achieve throughput or performance improvements
equivalent to the ratios stated here.
Photographs shown may be engineering prototypes. Changes may be incorporated in production models.
28
Thank You
29