Skip to main content

Full text of "memorex :: disc :: 3680 :: 102745034"

See other formats


iii 


HOW THE NEW 
MEMOREX 3680 
MEASURES UP. 


A DP MANAGER'S GUIDE 
TO DISC DRIVE 
PERFORMANCE EVALUATIONS. 


7 
. Y m nf 
: / 7 Val 
ly) Mi Ss ae” 


\ 0 ot 
antl 3 - 


ou 
\ = 


wy r | eae (i ! 
cae J y0% 3 en) oul ie 


@ 
/ 























ri 
/ Pn, 











g 





/ 





noitoubon! 

juqn uot bsvoigmi 1ot ses0 SAT 
2boriiem ee9oos sisb IsonoteiH 

ebortiom oninisq O8EE 

eborism pninisq 0808 

0898 .ev O8EE :enoeiisqmoo »*ismrionsa 
bores neito t2zom enoitesup onT 

Soren mort op voy ob sisiW 


atouborg sidiisgqmoo pulg yllsoinoteiH 

bsonsring need svsd xe1omeM yd beqoleveb 
.emaleye pniteixe of fosqesi ritiw ylds1abienoo 
miit nitt to Insvbs srit xevewor ,ebnim yasm nl 
-muotio to tse wen s betss1o esd ypolonrlost 
nint pniqolsveb to esitluaittib ort naviD .esonsie 
of Isoipol fi 2i ,sosiq tetit ortt ni egitilidsqso milit 
(MOQ) 1e1wiostunsM slditsqmoO gul4 s fosqxe 
Soniietto Isitini tint ritiw his srit sonsvbs oels of 


ynisgmoo sri no ebnegsb ti yleuoivdO 
even ,glqmsxe tot emeteye miit nint xs1omeM 
2s bnA .VYer sonia insmgolsvsb 1sbnu need 
,emaieve bsen miit nirlt to esiige s ni bnoose ont 
\ebienoo etosliay ybssils O88€ xsiomsM srt 
.noitulovs toubo1g ids 


rslensit sisb \tiosqso spsioie to enmisi nl 
art ot Isoitnabi 2i ti ebseqe eesoos bns 981 
noitizog \Winsg eit sono xevawoH .msieye O8EE 
orlt to erenpiesb srit ,berlose1 esw O88 rit dtiw 
atnemsonenns of noiineiis tisrit bsmut O86E 
\llsoifsmisib tuqdpuornt svoigmi bluow isnt 
pin o1s insmevsinos ns cove rot eorsie oT 
avs nso inernevorgmi tesborm s neve seusosd 
.2192U 1ot esonsupsenos pnirleinoies 


On Ga = (GON i) Syren sr. 


stances. Given the difficulties of developing thin 
film capabilities in the first place, is it logical to 

expect a Plug Compatible Manufacturer (PCM) 
to also advance the art with their initial offering? 


Obviously it depends on the company. 
Memorex thin film systems, for example, have 
been under development since 1977. And as 
the second in a series of thin film head systems, 
the Memorex 3680 already reflects consider- 
able product evolution. 


In terms of storage capacity, data transfer 
rate and access speeds, it is identical to the 
3380 system. However, once this parity position 
with the 3380 was reached, the designers of the 
3680 turned their attention to enhancements 
that would improve throughput dramatically. 
The stakes for such an achievement are high, 
because even a modest improvement can have 
astonishing consequences for users. 


Benchmark comparisons: 3380 vs. 3680 
The questions most often asked 
Where do you go from here? 


Introduction 1 Historically, plug compatible products 
The case for improved throughput 1 developed by Memorex have been enhanced 
Historical data access methods { considerably with respect to existing systems. 
3380 pathing methods 9 Inmany minds, however, the advent of thin film 
3680 pathing methods 3 technology has created a new set of circum- 

4 

6 

8 


A study of interactive user productivity by 
IBM's Arvind Thadhani’ identified a significant 
increase in productivity as system response time 
decreased from five seconds to one second. 
Moreover, as system response time dropped 
into the subsecond range, the increase in user 
productivity skyrocketed (see Figure 1). In fact, 
in the range between 3.0 seconds and 0.3 
seconds, user productivity increased almost 
three times as fast as the system response time 
decreased. 


500 


400 


8 


TRANSACTIONS 


8 





0! 
1 2 
SYSTEM RESPONSE TIME 
(SECONDS) 


Figure 1: Plot of System Response Time 
vs. User Transaction. 


The implications in terms of bottom line 
benefits are impossible to ignore. In one engi- 
neering study, for example, users working at 
on-line graphics terminals with a light pen 
improved productivity by as much as 550% 
when system response time decreased from 
1.5 seconds to 0.4 seconds. 


In still another study involving code gener- 
ation, a test project was completed in just 61% 
of the time projected when system response 
time was decreased from 2.22 seconds to 0.84 
seconds. What's more, there was a 57% reduc- 
tion in the error rate. 


Clearly, if user productivity increases 2 to 5 
times, the cost savings quickly soar to significant 
proportions, but there are other ways to capital- 
ize on throughput improvement. You could, for 
example, add more terminals without degrading 
your present system response time. And you 
could obviously improve the turnaround time 
on your batch processing. Scheduling also 





improves because even peak loads can be 
dispatched within predictable and consistent 
time frames. All told, the effect on DP opera- 
tions is similar to adding a more powerful 
processor ...an expensive eventuality most 
companies would prefer to defer as long as 
possible. 


Historical data storage access 
architecture. 


50 
45 
40 
35 
30 


25 


RESPONSE TIME IN MILLISECONDS 


20 


15 
30° 60 90 120 150 180 


SUBSYSTEM I/O OPERATIONS PER SECOND 
Figure 2: Plot of I/O's vs. Time. 


As processing power dramatically increased 
the number of I/O's executed, accessing the 
burgeoning quantities of data continued to be 
possible only at the expense of processing 
speed. The fundamental problem was simply 
that the I/O's couldn't get out of their own way. 
(Figure 2.) 


Various access architectures were developed 
in an effort to improve the throughput. The main 
objective was to prevent a single I/O from 
blocking access to the entire string as it obvi- 
ously would do if all the actuators were simply 
connected in series. Thus, there were solutions 
that provided multiple paths, various path 
switching arrangements, parallel-series hookup 
arrangements and so forth. Some worked 
better than others but all of them improved data 
access. 


In 3380 class subsystems, the sophistication 
of the data access architecture is a constant 
reminder of the importance of the problem. 


™FASTER IS BETTER —A Business Case For 
Subsecond Response Time; COMPUTERWORLD, 
April 18, 1983; Vol. 37, No. 16. 





The IBM 3380 Subsystem utilizes a 
configuration called Dynamic Path Selection 
(DPS) whereby groups of four actuators share 
an internal path. (Figure 3.) The four internal 
paths are connected to the two string controllers 
through a “Two by Four Switch” which allows 
the paths to be connected to either string 
controller 


In order to establish a connection between 
a Storage Director, a String Controller and an 
actuator, it is necessary not only that these be 
free, but necessary also that no other actuator 
on the same internal path be communicating 
with the other string controller. This additional 
requirement decreases the probability of access 
and contributes to slower response because 
communications with any one actuator effec- 
tively ties up 2,520 megabytes of data, the 
storage capacity of the four actuators on 
that path. 


To minimize this problem, the 3380 is 
frequently configured as two half strings. This 
improves the overall response time compared 
with a full string because it reduces the conten- 
tion for internal paths. This solution has a 
downside, however, in that each string, whether Figure 3: 3380 Pathing Method. 
short or long, requires an expensive string 
controller. And for every two string controllers, 
an equally expensive storage controller is 
required. 


3880-3 
STORAGE 
CONTROL 


3380 AA4 










3380 B4 


3380 B4 





3380 B4 


3888 DUAL 


3683 DUAL 
STRING 
CONTROLLER 






3680 DISK MODULE 


3680 DISK MODULE 





3680 DISK MODULE 





Figure 4: 3680 Pathing Method. 


The Memorex 3680 Subsystem (Figure 4.) 
employs a pathing arrangement called “Maxi- 
mum Availability Path Selection” or “MAPS” 
whereby each actuator is dual pathed and totally 
independent from each other. Together with 
separate actuator microprocessors, this enables 
simultaneous data transfer from any two actua- 
tors within a string, including to and from the 
same HDA. 


Since the two data paths have the ability to 
access all of the actuators within the string, the 
data availability of the 3680 is far superior to 
that of the 3380, whether it is configured as a 
half string or a full string. In fact, our DASD 
Performance Analytical Model shows that a 
3680 full string will outperform a 3380 half string 
simply because the maximum unit of contention 
in a 3680 string is only 630 megabytes — the 
amount of data stored under a single actuator. 





This same model currently in use also 
predicts that MAPS will provide I/O transaction 
rates comparable to those of a 3380 configured 
with the Dynamic Reconnect feature of 
MVS/XA. In fact, when 3680's are optimized 
for use with MVS/XA, our models predict they 
will be able to use Dynamic Reconnect to even 
greater advantage so that the improvement 
already achieved will be even greater. 


The MAPS feature also incorporates an 
improved protocol capability, which enables 
more data to be transferred per control opera- 
tion. For example, MAPS transfers seek and 
set sector information in one control operation 
and does not need additional control operations 
to verify the transfer. With the 3380 system, 
additional control operations are required to 
verify seek and set sector and to verify the 
parameters transferred from these first two 
operations. 


Further improvement in throughput results 
by staggering a track’s index mark 180 degrees 
apart on adjacent cylinders. This speeds 
sequential processing because less time is 
required for latency. The result is that data can 
be retrieved faster because the system can 
start looking for it sooner. 





Described in the following are the initial 
benchmark testing results which compare 
3680 subsystem performance with that of the 
3380 subsystem. 


Figure 5 summarizes the results of a 
benchmark run for a large U.S. customer. The 
benchmark jobstream consisted of a program 
which executed job steps writing sequential 
records using multiple blocksizes. The program 
then did random reads using the same block- 
sizes and data sets created by the sequential 
writes. 


Multiple jobstreams were run against multiple 
actuators on identical configuration 3680 and 
3380 subsystems. These jobstreams were 
executed on a 3083, model E16 processor 
complex controlled by an MVS operating 
system. 


The 4x4 test demonstrates the comparison 
found on identical logical volumes when four 
copies of the jobstream were run against two 
HDAs or four actuators on both the Memorex 
3680 and IBM 3380 configurations. In the 3380 
case, each HDA (and thus its two actuators) 
was on a different Dynamic Path Selection 
(DPS) internal path, so the unit of contention 
was kept to the minimum 1260 megabytes. 
The 2x2 tests demonstrate the comparison 
found on identical logical volumes when two 
copies of the jobstream were run against 
one HDA. 


The 4x4 test constitutes the best case for the 
3380, in that the HDA\‘s were on different DPS 
internal paths. However, by using this configura- 
tion an improvement close to what may typi- 
cally be expected is produced. The 2x2 test 
constitutes the worse case for the 3380 in that 











the two actuators were on the same DPS inter- 
nal path, while the Memorex case could fully 
utilize the unrestricted dual paths provided 

by MAPS. 


You can see a further example of the MAPS 
benefit in Figure 6. It summarizes results of a 
program which provides the ability to select |/O 
transaction rates, block sizes, read or write 
operations, and access modes. This was exe- 
cuted on the same processor with the same 
software as the preceding benchmark, but on 
Memorex and IBM strings with 4 spindles and 
8 actuators. 


PROGRAM THROUGHPUT IMPROVEMENT 
(MRX TO IBM) 








90 


80 


70 


60 


50 


40 





30 


20 


ELAPSED TIME (MINUTES/SECONDS) 


10 


TIME 
DIFFERENCE | 22:34 | 21:06 | 18:52 | 16:46 | 30:56 | 32:32 | 


A% FASTER 
TEST 














| MEMOREX 
@ IBM 


Figure 5: Benchmark Comparisons. 


The results show that the throughput and 
performance improvements are sustained 
throughout common (4K and 6K) and less 
common (1K and 2k) blocksizes. They also 
show the improvements increase as |/O 
rates grow. 


The most apparent observation that may 
be made with this benchmark is the Memorex 
subsystems ability to handle significantly higher 
\/O transaction rates. This will provide benefits 
in peak transaction periods and when I/O's are 
skewed to a few actuators in the subsystem. 
It will also enable the Memorex subsystem to 
maintain a more stable transaction time ata 
terminal or turnaround for batch workload. 





For termina 
important, a: 


As you ca 





response time this is particularly 
s we have earlier discussed. 


nsee from these results, the 


3680 disc subsystem can significantly outper- 
form the 3380 disc subsystem. 


Performa 
being made 


nce comparisons are continuously 
by Memorex; if you would like to 





receive copi 


es of future benchmark reports, 


just write “Benchmarks” on your business card 





and mai 
the brochur 


litto us at the address on the back of 


e. 


1/0 THROUGHPUT IMPROVEMENT 





AVERAGE 


INCREASE +42% 


200 


180 


160 


1/O PER SECOND 


80 


60 


40 


20 





10) 


(VO RESPONSE)}| 15 


16 17 


(MRX TO IBM) 


+35% 


+40% +42% 








19 20 21 16 26 





BLOCK 
SIZE 





1K RANDOM 


G@ MEMOREX 
B IBM 


Figure 6: Benchmark Comparisons. 


2K RANDOM 





4K RANDOM 


6K SEQUENTIAL 








Q: Do | need additional or special software for 
MAPS? 

A: No. MAPS is transparent to the operating 
system and will be supported by any oper- 
ating system supporting 3380 subsystems. 


: What is Dual Path? 

: Dual Path is a feature which provides a 
second path to each actuator within the 
string, permitting simultaneous read/write 
operations on any two different actuators 
within the string — even on the same Head 
Disc Assembly (HDA). 


PO 


Q: How does it improve throughput 
performance? 

A: Each of the two data paths can control any 
actuator in the string by means of the 


This feature improves data availability by 
reducing the Unit of Contention, False 
Device or Actuator Busy and missed RPS 
interrupts. Throughput is further facilitated 
by an improved protocol that speeds com- 
munications between components of 

the system. 





Q: What is the Unit of Contention? 

A: The amount of data that becomes unavail- 
able due to blocking by a data transfer at 
any one actuator. With the 3380 2x4 switch, 
one actuator can block out 3 others at 630 
MB each for a total of 2,520 MB. On the 
3680, the Unit of Contention is only 630 MB 
(the actuator in use) because alternate data 
paths solve the blocking problem. 


aximum Availability Path Selection (MAPS). 


Q: What is False Device Busy? 

A: False Device or Actuator Busy is a condition 
where a component constituting the path to 
the device or actuator is busy and cannot be 
recognized by the I/O control software. In 
this situation, it falsely appears that the device 
or actuator is busy. If, for example, an internal 
path of a 3380 is busy because a particular 
actuator is in use, attempts to utilize other 
actuators on the same path will result in a 
false busy. 


Q: What is a Missed RPS Interrupt? 

A: A Rotational Position Sensing Interrupt 
signals that a desired data location is about 
to pass beneath the read/write head. If the 
data location cannot be accessed due to 
the blocking problem discussed earlier, a 
Missed RPS Interrupt occurs. This intro- 
duces a delay of at least 16.6 ms which is 
the time required for an additional 360° 
rotation of the disc. 


Q: What is Dynamic Reconnect? 

A: Dynamic Reconnect is a feature of the 
MVS/XA operating system which uses hard- 
ware in the channel subsystem to provide 
a different reconnect path back to the 
processor. 


Q: Will MAPS support Dynamic Reconnect? 

A: Yes. Current 3680's offer comparable 
throughput and performance to 3380's with 
Dynamic Reconnect; when Memorex 3680 
Dynamic Reconnect is implemented, the 
improvement will be dramatic. 


Q: What is a 3380 internal path? 
A: A path within a 3380 string to which up to 
four actuators may be attached serially. 


Q: What is Extended Architecture? 

A: Extended Architecture (XA) is a new hard- 
ware architecture developed by IBM that 
provides 31-bit addressing and moved |/O 
processing from the CPU to an External 
Data Controller (EXDC). 


: Is there any internal path queueing? 

: No. I/O queues are not formed for the 
internal paths of a 3380 string. Should an I/O 
be issued for an actuator on an internal path 
already busy, a false busy will occur. 


Q: Does MAPS alter the ratio of 3680's required 
to replace 3350's? 

A: Yes. For the 3380, IBM at times has recom- 
mended a conversion ratio of 1.5 to 1; 
however, the 3680 ratio is 2 to 1. This is due 
to the throughput improvement of the 3680 
which permits greater effective use of the 
storage capacity. 


>o 


Q: Does block size affect the performance of 
MAPS? 

A: Yes. The improved protocol of MAPS 
reduces the overhead associated with every 
|/O operation. Therefore should less effective 
smaller block sizes have to be used with 
3380 technology, the 3680 benefit will 
increase. 


Q: |s path utilization affected by MAPS? 

A: Yes. Due to the enhanced protocol, utiliza- 
tion of components that constitute a path is 
reduced. Therefore, for the same amount of 
path utilization, the number of I/O's will be 
greater. 


Q: What channel queueing algorithm should 
| use for MAPS? 

A: To insure both paths are utilized effectively 
with a single processor, the Channel Rotate 
Balanced Scheduling algorithm should be 
used. When multiple processors are 
attached to the 3888, then the Last Channel 
Used Scheduling algorithm should be used. 


Q: Do other PCM’s have the equivalent of 
MAPS? 

A: Not really. Other PCM'’s offer Dual Port 
capability which share similarities with 
MAPS, however the improved protocoling 
of MAPS together with staggered index 
marks improves the throughput potential 
considerably. 


.../S to prove it to yourself. In addition to 
the conclusions reached via benchmarks, it is 
possible to predict with considerable accuracy 
the improvement in throughput performance 
for any given DP operation. Using an advanced 
analytical model, Memorex Systems Engineers 
can perform on-site evaluations and provide 
you with documented projections, typically in 
one day. 


An even more exacting evaluation can be 
performed in Santa Clara, California, using 
your software to develop comparisons on both 
3380 and 3680 subsystems. 


We invite you to be amazed. 





of noifibbs nl tlsewoy of ti svorg of ai... 
2i ti ,exsmnoned siv berlossy enoieulonoo eri 
yosiuoos sidsiebienoo miiw foibsig ot sldizzoq 
sonsnmmonheg tuqn quoi ni insmavoigmi srit 
bsonsvbs ns pnieU .noiisisqo 9 nevip yns tot 
21eeniona emeteye xe1omeM ,lsbom Isoityisns 
gbivorg bnis enoitsulsve stie-no mioheq nso 
ni vllsoiqyt enoifogjo1g bsingsmusob fiw voy 
ysb sno 

9d nso noifsulsve pnitosxe siom neve nA 
onieu ,simotilsO ,s1siO sinse ni bsrmoheq 
riod no enoziisqmoo goleveb ot siswitoe woy 
emsieyedue O88 bns O8EE 


.bossms 9d of voy stivni SW 




















Memorex Corporation 


San Tomas at Central Expressway 
Santa Clara, California 95052 
Telephone: (408) 987-2301 


When it matters, make it Memorex" 


MEMOREX