| Review of the I/O module...
1-4a, "materials" --
1 hours?? No way. This module easily takes me 2-3 hours, maybe more.
1-4, "<REFERENCE>"
Lots of these throughout the chapter...I figure you will find them all.
Also, last bullet is 2 bullets in one!
1-5, "Buffer" --
generic terms are not normally capitalized like BUFFER & TRACK
1-6, "OpenVMS" --
This is used frequently. Are we going to make the appropriate notation
when something is specifically VAX/VMS vs. Alpha/VMS?
1-7, 2nd bullet --
Note that there are a few old Direct I/O Printer Drivers. These are
rare but worthy of noting that it is the HW/SW developers that decide
which method of I/O is used. Also, mention terminals since it is usually
the biggest used of Buffered I/O is many systems.
1-8a --
Good list, but don't forget to mention NPAGEDYN and xRP-lists since this
resource is the 'transport' for Buffered I/Os. Raising the parameters on
this page increases pool consumption.
1-8 -- Ditto
1-9a --
On page 1-8a pql_Mxxxx values were listed. To be consistent let's get them
here as well. Also, don't forget that Direct I/O is mapped to the working
set, consuming user memory. A note here that increasing DIOLM may *NOT*
improve things if process memory is restricted -- when you increase DIOLM
significantly, also increase working set values, specifically WSdef and
WSquo.
1-9, 1st bullet --
"...user's address space." is actually working set. Let's mention it.
1-10 --
Also mention that the Control-T function shows combined Buffer and Direct
I/O on the 1-line status display.
1-11a --
I know it says numbers will be discussed later, but I didn't find them. We
need to give them ball parks to watch. Rule of thumb is the BIO count's
threshold is *approximately* the page fault threshold, i.e. VUP * 100 to
get a start. If the value is greater than this it is worth investigating.
Ditto with DIOs, but the formula is VUP * 30.
1-11, date --
Many of the display have older dates...can we update them? All of them!?
1-12, no mention of DECram Ram disk software.
1-14, "System Interconnection" --
Kinda bare! Very vague! Is it a cable? A rack? Is it SW or HW? Is it
the same thing as a 'backplane'? We need a brief description of it on the
instructor's page at least.
last bullet --
Drop it or put it on the Instructor's page. Not worth a lecture point.
1-15, bullets 4-6 --
short examples of each, KDB50, ISE(RFxx), HSC60
1-16a --
1st paragraph --
"At present these disks only allow data to BE written once." (Missing
the 'BE'). Also, not that this is true for DEC hardware...3rd parties
have some limited implementations of WMRM (write many) optical disks
already.
By the way, if it's too slow for consideration why do we mention it on
page 1-16?
2nd & 3rd paragraphs (sentences?) --
removEable doesn't need the extra 'E'.
3rd paragraph (sentence) --
Why say we aren't considering removable disks when on 1-18a (and others?)
we mention RA60s -- removable disks.
Last line --
"Open VMS" doesn't need the extra space.
1-16 --
bullets 1-4 --
Why not mention DECram (ram disks in general) and CDrom disks?
5th bullet --
It is? How? This is a leading statement maybe best left for the
instructor's page.
7th bullet --
"...does not touch the disk." -- not true for floppies, the head actually
rides on the media.
Last bullet --
"(but not always)" When is it NOT in a block (or sector)? Why would we
be interested in this?
1-17a, 4th item --
I thought Transfer time was getting the data from the disk, through the
head, through the controller and out on the bus. Is Transfer time *just*
the time 'skimming' over the data?
1-17, diagram --
It should be noted that most disks today use the top of the top platter and
some use the bottom of the bottom platter. This diagram is indicative of
older removable media.
1-18a --
There is a reference to the Quorum magazine, but no date or volume number!
Also, the list of disks is pretty old (We don't currently manufacture any
of those) and most new students have the newer/better drives -- RF73, RA72,
RA92, etc.
And the info at the bottom regarding the ESE20 needs updating. The ESE50
was written up in the current Sales Update and in Quorum for May 1992.
1-18 --
This information isn't organized correctly. Let's define the parts of the
I/O process and THEN talk about the performance issues. Take the first 3
bullets (first 1/2 of page) and put them on the bottom of 1-20 or 1-21.
Also, the bottom half belongs probably at the bottom of 1-21 or 1-22
1-19a, lst line --
belongs on page 1-21a
1-20a --
An important note is that many disk task are heavily overlapping so there
is optimal usage of time.
1-21a --
Yes, HSCs can have the largest Controller Latency, but with their optimizing
features they can handle a heavier work load more efficiently, thus with
increased workload the overhead there can reduce Seek and Rotational latency.
1-21 --
If I recall correctly this was also done with an RP06 on a MASSBUS, right?
1-22a, middle --
"An HSC may contain up to 8 data channels, DEPENDING ON THE MODEL."
1-22, last topic --
No mention of DSSI controllers?
1-24a --
1st sentence --
"I.O" should be I/O.
3rd bullet --
"25/" has extraneous "/"
4th item from bottom --
These numbers seem very old, like RA81 data. Can we get more current
data for RA92, RF73 and RA72?
1-24, 4th bullet, dashes --
This is *not* to say that numbers outside this boundary is bad, merely
it is time to look at the devices.
1-26 --
2nd bullet --
"...the top *direct* I/O users." -- add DIRECT
7th bullet --
the concept of Primary page files need to be cleared up. Back in V4.x
there was a primary page file. There really isn't such as beast anymore.
Certain items *HAD* to page to the older Primary file, but with V5.x
any application can page to any file. By convention we think of the
page files on the system disk as Primary, but this is no longer true.
Now, we generally want the page files to be approximately evenly
balanced. A process spreads out its I/Os across as many as 4 page files
per system -- EVENLY! This mean that if a process needed to page out
100000 blocks and the PRIMARY had 40k and the SECONDARY had 60k, the
process we end up in PFW/RWMPB state because it would have 40k in each
file and be unable to us the remaining 20k. Rule of thumb...keep page
files approximately evenly sized on the quietest/fast media that you can.
$
|
| Continuing...
1-27a, 4th paragraph --
"Consistently seeing a nonempty value...", considering we are running at
priority 4, are we going to see anything useful? I've heard various
internals and device driver instructors say this was a waste of time.
last paragraph --
"Note that the DPQFL and DPQBL both point to the head of the queue..."
HUH? Maybe I'm dense. Maybe I don't use SDA enough. But, I can't see
this. Care to point it out?
1-27 --
5th bullet, 1st dash, "...nine to eleven..." --
Hasn't this number changed? We aren't usually going to saturate a
HSC, maybe the HSC50, but not the others, rather the requestors will
saturate first almost 100% of the time.
HSC50's, 40's and 70's are mentioned. What about 60's and 90's?
Last bullet -
Cute, but not really useful. Move to instructor's page.
1-28, last bullet --
ditto, move to instructor's page. Average system manager doesn't
read SDA!
1-29a, 2nd paragraph --
typo, priMcipal
4th paragraph, 2nd line --
typo, "strip" needs an 'e', same with 5th paragraph
1-29, 2nd bullet --
'stripset' needs an 'e'
also, is Stripe Set 1 word or 2? Stripeset or Stripe set?
1-31a, 1st sentence --
typo, missing 'of' either before or after the parenthesis.
1-33a, 1st paragraph --
Bolded info regarding VAXft 3000 and the next sentence are old stuff.
as of V5.5 this isn't an issue. (Also, looks like it was supposed to
be it's own paragraph)
1-33, dots or dashes?
shouldn't the indented 'dots' be 'dashes'?
1-36 --
First sentence -- shouldn't File Control Processor be in 'caps'?
2nd bullet -- "OPen" has capitalized 'p'. Also, isn't RENAME an XQP
function?
13th bullet -- Are the tense right? Are there a problem with Are/Is in
this line? Also, is the content 'undefined' or 'indeterminate'?
1-38a --
These comments belong on page 1-39a
1-38, last two lines --
Is VAX RMS information supposed to be a bullet? Shouldn't there be a
bit of text following retrieval pointers?
1-41, last lines --
we should mention defrag utilities, especially since we now make one!
1-42a --
You overlooked the fact that we have on-line defragment utility.
You also overlooked that we are working with the various vendors to
make sure they understand how MOVEFILE works and to encourage their
use of it. This memo looks old. We announced the defrag utility over
a year ago in Atlanta DECUS!
Last paragraph --
There are numerous CADCAM applications that require that their large
datafiles are contiguous -- no feasible way other that defragers. There
are valid reasons, HOWEVER, defragers are the single most *over-sold*
utility on VMS. Over 75-80% of system managers admit that they really
don't need it or the performance gain is minimal (Digital News, if I
can find the article I'll post the volume number).
1-43a, last line --
"...it might be time to DEFRAGMENT disks." Backup is NOT the issue,
defragging is. How they do it is there problem.
1-44a, chicken --
How did a 'hen' get into the paragraph that starts with "See also..."
maybe it is 'when'?
1-45, 2nd and 3rd bullets --
This is not how I understand it. Highwater marking has changed from
it's old 'erase on allocate' action to 'erase on access' activity as of
V5.0. That is unless V5.5 introduced something I didn't hear about.
Maybe you were thinking of erase-on-delete which is set with its own
qualifier.
1-47a, last paragraph, 2nd sentence --
reads funny, are we missing a word or two?
3rd bullet --
looks totally out of place compared to the other bullets.
$
|
| And to finish up....
1-44, 1-47 (& 1-48) --
These pages belong beside each other. As we talk about the caches lets
put the Monitor command near the text.
1-48a, last lines --
Looks like these Bolded words were supposed to be bullets or a vertical
list.
1-52a, last sentence --
"...paged to any PAGE file" - misleading. Add the word PAGE.
also, what are the last two bolded items supposed to do?
1-52, 'buckets' --
We might want a short explanation of Buckets on the instructor's page.
Also, read the title of the page followed by the next two bullets. Seems
to be missing some some words, as it doesn't make sense.
1-57a, 2nd sentence --
typo, "...is limited in the..." maybe "is" instead of "in"?
1-57, commas missing...
"Terminal Operations, when improperly handled, can..."
Also, why is "Operations" in Caps?
1-58a, 2nd sentence --
Add the following line after the sentence on the DZ-11...
"The DZ-11, like the VAX 11/780, is a standard against which many I/O
devices are measured."
1-59a, 1st sentence --
"asymmetric" is not clear. What is intended statement?
1-62, 5th bullet, 2nd dash --
...or SET LOGIN/INTERACTIVE=nn
Also, I noticed you took out both graphs with the DMA and the graphs of
LAT vs. DZ vs. DMF. These were useful to illustrate the impact on the CPU
in regards to the newer devices. Please bring them back as appendix pages,
instructor's pages or even back to the student page.
1-63, 3rd bullet --
For the System Manager's purposes, buffer sizes are almost never changed.
The big change is in the quantity of buffers (which is mentioned later).
Same for sentence following 4th bullet, "...by maintaining the size AND
NUMBER of...
1-64a, last paragraph --
V4.6 and V4.7 were left out. These were capable of RECEIVING packets out of
order, but not TRANSMITTING them out of order.
1-64, 4th bullet --
add the following to the end of the sentence, "... , although NETCONFIG
assigns default values."
NCP text on bottom of page --
To make this stick these need to DEFINEd as well as SET. Also, verify that
path splitting is enabled: "NCP> DEFINE EXECUTOR PATH SPLIT POLICY NORMAL"
(and SET).
1-65a, need text --
describe the relationship between Transmit and Receive buffers.
1-65, 2nd set of bullets --
4th and 5th bullets -- these settings are handled by NETCONFIG.
7th bullet --
is VERY common on Pathworks host systems, Local Area clusters,
major routing hubs...MUCH more common than changing the buffer size.
1-66a, last line --
4 SYSGEN parameters jammed at end of sentence...are they supposed to be
bullets?
1-66, 1st bullet --
LRPCOUNT is now calculated by feedback, LRPSIZE is still based on HW.
On to the VAXcluster chapter...
$
|