Skip to contentSkip to conversation

Difference in priority

DISCUSSION Albert Cahalan 4 participants 7 messages Oct 12, 2004 – Oct 13, 2004

lore
Share & tools
lore
#0 lore raw
Quoted 11 lines · Con Kolivas
Quoted 8 lines · Ankit Jain

if somebody knows the difference b/w /PRI of both these commands because both give different results

ps -Al
& top

as per priority rule we can set priority upto 0-99 but top never shows this high priority

Quoted 6 lines

Priority values 0-99 are real time ones and 100-139 are normal scheduling ones. RT scheduling does not change dynamic priority while running wheras normal scheduling does (between 100-139). top shows the value of the current dynamic priority in the PRI column as the current dynamic priority-100. If you have a real time task in top it shows as a -ve value. ps -Al seems to show the current dynamic priority+60.

What would you like to see? There are numerous competing ideas of reality. There's also the matter of history and standards. I'd gladly "fix" ps, if people could agree on what "fix" would mean.

Various desirable but conflicting traits include:

a. for normal idle processes, PRI matches NI b. for RT processes, PRI matches RT priority c. for RT processes, PRI is negative of RT priority d. PRI is the unmodified value seen in /proc e. PRI is never negative
f. low PRI is low priority (SysV "pri" keyword) g. low PRI is high priority (UNIX "PRI", SysV "opri") h. PRI matches some kernel-internal value
i. PRI is in the range -99 to 999 (not too wide)

I had originally tried to match up with Solaris and a few other systems, but that's looking hopeless. I intend to change to something sane, perhaps even making the PRI of "-o pri" be the same as that of "-l". Currently I don't even try to document PRI.

It is good to keep the value narrow. I really wish we didn't have so many RT levels; POSIX only requires that there be 32. A simple 0..99 value would be great.

Here is what I have in my current code, with the headers edited so they won't be confusing. Note that the left two versions ("aaa" and "bbb") are similar to the traditional SysV "pri", but the new POSIX and UNIX standard makes them unsuitable for use as the "PRI" displayed by "ps -l". It is common to generate the "PRI" of "ps -l" via an "opri" keyword.

$ LD_LIBRARY_PATH=../../proc ../../ps/ps -t pts/8 --sort=-priority -o stat,pri,pri_api,rtprio,pri_bar,priority,pri_baz,opri,pri_foo,ni,cls,sched,comm
STAT aaa bbb RTPRIO ccc ddd eee fff ggg  NI CLS SCH COMMAND
SN     0 -40      -  40  39 139  99  19  19  TS   0 setpriority19
Ss+   23 -17      -  17  16 116  76  -4   0  TS   0 bash
S<    39  -1      -   1   0 100  60 -20 -20  TS   0 setpriority-20
S     41   1      1  -1  -2  98  58 -22   -  RR   2 min_rr
S    139  99     99 -99 -100  0 -40 -120  -  FF   1 max_fifo
#1 lore raw
Quoted 22 lines · Albert Cahalan
Quoted 5 lines · Con Kolivas
Quoted 3 lines · Ankit Jain

if somebody knows the difference b/w /PRI of both these commands because both give different results

Quoted 8 lines
Quoted 5 lines

ps -Al
& top

as per priority rule we can set priority upto 0-99 but top never shows this high priority

Priority values 0-99 are real time ones and 100-139 are normal scheduling ones. RT scheduling does not change dynamic priority while running wheras normal scheduling does (between 100-139). top shows the value of the current dynamic priority in the PRI column as the current dynamic priority-100. If you have a real time task in top it shows as a -ve value. ps -Al seems to show the current dynamic priority+60.

What would you like to see? There are numerous competing ideas of reality. There's also the matter of history and standards. I'd gladly "fix" ps, if people could agree on what "fix" would mean.

Various desirable but conflicting traits include:

Quoted 9 lines

a. for normal idle processes, PRI matches NI b. for RT processes, PRI matches RT priority c. for RT processes, PRI is negative of RT priority d. PRI is the unmodified value seen in /proc e. PRI is never negative
f. low PRI is low priority (SysV "pri" keyword) g. low PRI is high priority (UNIX "PRI", SysV "opri") h. PRI matches some kernel-internal value
i. PRI is in the range -99 to 999 (not too wide)

I can't say I've ever felt strongly about it. Wish I knew what was the best way. If we change the range of RT priority range by increasing it from 100 to say 1000 then any arbitrary value to subtract will be wrong. How about just leaving the absolute dynamic priority value? Then we don't have any negative values confusing it, it isn't affected by increasing the range of RT priorities, and better priority values still are lower in value.

Cheers,
Con

#2 lore raw
Quoted 38 lines · Con Kolivas
Quoted 26 lines · Albert Cahalan
Quoted 11 lines · Con Kolivas
Quoted 8 lines · Ankit Jain

if somebody knows the difference b/w /PRI of both these commands because both give different results

ps -Al
& top

as per priority rule we can set priority upto 0-99 but top never shows this high priority

Quoted 6 lines

Priority values 0-99 are real time ones and 100-139 are normal scheduling ones. RT scheduling does not change dynamic priority while running wheras normal scheduling does (between 100-139). top shows the value of the current dynamic priority in the PRI column as the current dynamic priority-100. If you have a real time task in top it shows as a -ve value. ps -Al seems to show the current dynamic priority+60.

What would you like to see? There are numerous competing ideas of reality. There's also the matter of history and standards. I'd gladly "fix" ps, if people could agree on what "fix" would mean.

Various desirable but conflicting traits include:

Quoted 9 lines

a. for normal idle processes, PRI matches NI b. for RT processes, PRI matches RT priority c. for RT processes, PRI is negative of RT priority d. PRI is the unmodified value seen in /proc e. PRI is never negative
f. low PRI is low priority (SysV "pri" keyword) g. low PRI is high priority (UNIX "PRI", SysV "opri") h. PRI matches some kernel-internal value
i. PRI is in the range -99 to 999 (not too wide)

Quoted 6 lines

I can't say I've ever felt strongly about it. Wish I knew what was the best way. If we change the range of RT priority range by increasing it from 100 to say 1000 then any arbitrary value to subtract will be wrong. How about just leaving the absolute dynamic priority value? Then we don't have any negative values confusing it, it isn't affected by increasing the range of RT priorities, and better priority values still are lower in value.

That's not convincing enough to overcome inertia.

I can't see why the RT priority range would be increased. It's overkill already, especially since Linux doesn't have priority inheritance. Since POSIX requires 32 levels, that is the right number. Actually using more than one level (remember: NO priority inheritance) might not be wise. If we stuck to 32 levels, RT and non-RT could both fit within a 2-digit positive number. Also, there'd be fewer bits for the scheduler to examine.

#3 lore raw
Quoted 39 lines · Albert Cahalan
Quoted 34 lines · Con Kolivas
Quoted 22 lines · Albert Cahalan
Quoted 5 lines · Con Kolivas
Quoted 3 lines · Ankit Jain

if somebody knows the difference b/w /PRI of both these commands because both give different results

Quoted 8 lines
Quoted 5 lines

ps -Al
& top

as per priority rule we can set priority upto 0-99 but top never shows this high priority

Priority values 0-99 are real time ones and 100-139 are normal scheduling ones. RT scheduling does not change dynamic priority while running wheras normal scheduling does (between 100-139). top shows the value of the current dynamic priority in the PRI column as the current dynamic priority-100. If you have a real time task in top it shows as a -ve value. ps -Al seems to show the current dynamic priority+60.

What would you like to see? There are numerous competing ideas of reality. There's also the matter of history and standards. I'd gladly "fix" ps, if people could agree on what "fix" would mean.

Various desirable but conflicting traits include:

Quoted 9 lines

a. for normal idle processes, PRI matches NI b. for RT processes, PRI matches RT priority c. for RT processes, PRI is negative of RT priority d. PRI is the unmodified value seen in /proc e. PRI is never negative
f. low PRI is low priority (SysV "pri" keyword) g. low PRI is high priority (UNIX "PRI", SysV "opri") h. PRI matches some kernel-internal value
i. PRI is in the range -99 to 999 (not too wide)

Quoted 1 line

I can't say I've ever felt strongly about it. Wish I knew what was the best way. If we change the range of RT priority range by increasing it from 100 to say 1000 then any arbitrary value to subtract will be wrong. How about just leaving the absolute dynamic priority value? Then we don't have any negative values confusing it, it isn't affected by increasing the range of RT priorities, and better priority values still are lower in value.

That's not convincing enough to overcome inertia.

Quoted 8 lines

I can't see why the RT priority range would be increased. It's overkill already, especially since Linux doesn't have priority inheritance. Since POSIX requires 32 levels, that is the right number. Actually using more than one level (remember: NO priority inheritance) might not be wise. If we stuck to 32 levels, RT and non-RT could both fit within a 2-digit positive number. Also, there'd be fewer bits for the scheduler to examine.

As I said; I don't feel strongly. There was an option to change the range of real time priorities in 2.4 which is why I mention it. Let others speak cause I care not what you do as long as we understand it. I'm not changing the scheduler to suit userspace tools.

Cheers,
Con

#4 lore raw
In reply to Albert Cahalan
Quoted 5 lines · Albert Cahalan

I can't see why the RT priority range would be increased. It's overkill already, especially since Linux doesn't have priority inheritance. Since POSIX requires 32 levels, that is the right number. Actually using more than one level (remember: NO priority inheritance) might not be wise.

Linux will probably have priority inheritance soon. See the "Real Time Kernel" thread.

Lee

#5 lore raw
Quoted 11 lines · Lee Revell
Quoted 5 lines · Albert Cahalan

I can't see why the RT priority range would be increased. It's overkill already, especially since Linux doesn't have priority inheritance. Since POSIX requires 32 levels, that is the right number. Actually using more than one level (remember: NO priority inheritance) might not be wise.

Linux will probably have priority inheritance soon. See the "Real Time Kernel" thread.

Is that opinion based any on this article and Linus's comments in it?

http://news.com.com/A+new+direction+for+Linux+for+gadgets/2100-7344_3-5406291.html?tag=cd.top

Signature · 1 line
~Randy
#6 lore raw
Quoted 10 lines · Randy.Dunlap
Quoted 9 lines · Lee Revell
Quoted 5 lines · Albert Cahalan

I can't see why the RT priority range would be increased. It's overkill already, especially since Linux doesn't have priority inheritance. Since POSIX requires 32 levels, that is the right number. Actually using more than one level (remember: NO priority inheritance) might not be wise.

Quoted 7 lines
Quoted 2 lines

Linux will probably have priority inheritance soon. See the "Real Time Kernel" thread.

Is that opinion based any on this article and Linus's comments in it?

http://news.com.com/A+new+direction+for+Linux+for+gadgets/2100-7344_3-5406291.html?tag=cd.top

No, but priority inheritance is not the same as making Linux an RTOS.

Lee