D
Dan Irwin
I know this is a kind of stubid question, but what does the buffer on hard drive do?
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
R
Ruth
it buffers
ok...enough with the schmarty pants
Buffer is a place to hold data that comes from the drive, that may available
before the CPU is ready for it. If that happens, then when CPU is ready, if
the data isn't available, it has to call the drive up and say "hey you
remember that data i asked for a few nanoseconds ago, well, I need it now"
Without the buffer the drive grumbles and says oh #@#$!! and has to go fetch
it again. With the buffer, the data is stored in some RAM on the drive
electronics so that when the CPU says "OK" ready, the drive says "here you
are right in my hand" and doesn't have to go look again. The advantage is
that its a lot faster to get data from the buffer than from the disk
platters.
It works in a similar way for writing data. If the CPU can dump the data to
the drives buffer to be written when the drive is good and ready, it can say
"good, on to the next thing" instead of having to wait until the disk is
availalbe to perform the the write ....
Make sense. I hope so. At end of the day, more buffer is better for disk
intensive operations.
ok...enough with the schmarty pants
Buffer is a place to hold data that comes from the drive, that may available
before the CPU is ready for it. If that happens, then when CPU is ready, if
the data isn't available, it has to call the drive up and say "hey you
remember that data i asked for a few nanoseconds ago, well, I need it now"
Without the buffer the drive grumbles and says oh #@#$!! and has to go fetch
it again. With the buffer, the data is stored in some RAM on the drive
electronics so that when the CPU says "OK" ready, the drive says "here you
are right in my hand" and doesn't have to go look again. The advantage is
that its a lot faster to get data from the buffer than from the disk
platters.
It works in a similar way for writing data. If the CPU can dump the data to
the drives buffer to be written when the drive is good and ready, it can say
"good, on to the next thing" instead of having to wait until the disk is
availalbe to perform the the write ....
Make sense. I hope so. At end of the day, more buffer is better for disk
intensive operations.
drive do?Dan Irwin said:I know this is a kind of stubid question, but what does the buffer on hard
C
CJT
Ruth said:it buffers
ok...enough with the schmarty pants
Buffer is a place to hold data that comes from the drive, that may available
before the CPU is ready for it. If that happens, then when CPU is ready, if
the data isn't available, it has to call the drive up and say "hey you
remember that data i asked for a few nanoseconds ago, well, I need it now"
Without the buffer the drive grumbles and says oh #@#$!! and has to go fetch
it again. With the buffer, the data is stored in some RAM on the drive
electronics so that when the CPU says "OK" ready, the drive says "here you
are right in my hand" and doesn't have to go look again. The advantage is
that its a lot faster to get data from the buffer than from the disk
platters.
It works in a similar way for writing data. If the CPU can dump the data to
the drives buffer to be written when the drive is good and ready, it can say
"good, on to the next thing" instead of having to wait until the disk is
availalbe to perform the the write ....
Make sense. I hope so. At end of the day, more buffer is better for disk
intensive operations.
A couple of embellishments -
The drive can be clever about its read buffers, and pre-load into them
sectors that haven't yet been identified as desired, but which follow
ones that have been, under the assumption that data is often read
sequentially.
Write buffering has implications for error recovery -- what happens if
the power fails (or the machine reboots) while the drive's buffers are
full of information that has not yet been written to the disk? As a
result, it is often turned off in high reliability systems.
F
Folkert Rienstra
CJT said:A couple of embellishments -
Actually, it was total crap.
The drive can be clever about its read buffers, and pre-load into them
sectors that haven't yet been identified as desired, but which follow ones
that have been, under the assumption that data is often read sequentially.
They usually (or used to) call that cache.
Write buffering has implications for error recovery --
what happens if the power fails (or the machine reboots)
while the drive's buffers are full of information that has not yet been
written to the disk?
What if it is turned off and the drive's buffers are full of information
that has not yet been written to the disk?
The 'buffer' is 128kB, whether write buffering is on or off.
C
CJT
Folkert said:Actually, it was total crap.
"it?"
They usually (or used to) call that cache.
So?
What if it is turned off and the drive's buffers are full of information
that has not yet been written to the disk?
The 'buffer' is 128kB, whether write buffering is on or off.
In one case the computer to which it's attached thinks the write is
done.
D
Dan Irwin
Where does the ram play into this?
Also just so i have it right becacue of the way buffer functions it is
something that would not affect the seek time on a hard drive
Also just so i have it right becacue of the way buffer functions it is
something that would not affect the seek time on a hard drive
C
CJT
Dan said:Where does the ram play into this?
The computer also has buffers in RAM above and beyond those actually
on the drive.
It depends on what you mean. With enough buffers, the computer/driveAlso just so i have it right becacue of the way buffer functions it is
something that would not affect the seek time on a hard drive
combination can also be clever about the order in which pending accesses
are done, which can affect the apparent aggregate seek time (i.e.
avoiding having to seek, or shortening the seeks done, is as good as
brute force speeding up seeks). The devil is in the details. But I
would say the size of the buffers doesn't directly affect seek time.
F
Folkert Rienstra
To what, the crap?
The "it" that is presumably the same as "what".
Buffer is not the same as cache.
Which is the actual reason, to avoid that a system administers things as done
when it can't be sure of it and to ensure that IO that has to be done in a
certain order is actually done in that order, which is a requirement for certain
filesystems. In such a system you probably won't have queue reordering either.
But in those systems a battery backed-up cache controller is used that
reports the transaction completed immediately, as if it had been written.
"it?"
The "it" that is presumably the same as "what".
Buffer is not the same as cache.
In one case the computer to which it's attached thinks the write is done.
Which is the actual reason, to avoid that a system administers things as done
when it can't be sure of it and to ensure that IO that has to be done in a
certain order is actually done in that order, which is a requirement for certain
filesystems. In such a system you probably won't have queue reordering either.
But in those systems a battery backed-up cache controller is used that
reports the transaction completed immediately, as if it had been written.
D
Dan Irwin
So just to confirm a hard drive with a 8mb buffer and a 16mb buffer
could conceivably have the documented average seek time, while the
16mb one would be faster in real life application
could conceivably have the documented average seek time, while the
16mb one would be faster in real life application
C
CJT
Dan said:So just to confirm a hard drive with a 8mb buffer and a 16mb buffer
could conceivably have the documented average seek time, while the
16mb one would be faster in real life application
Could be.
I
Impmon
Write buffering has implications for error recovery -- what happens if
the power fails (or the machine reboots) while the drive's buffers are
full of information that has not yet been written to the disk? As a
result, it is often turned off in high reliability systems.
There used to be an issue with earlier version of XP, the PC would
shut down so fast the data in the buffer may not have been written
yet. A patch was released long ago to put a delay and force the drive
to finish writing before shutting down.
C
CJT
Impmon said:There used to be an issue with earlier version of XP, the PC would
shut down so fast the data in the buffer may not have been written
yet. A patch was released long ago to put a delay and force the drive
to finish writing before shutting down.
Thanks for reminding us of that, which is a similar issue. Of course,
a power failure won't allow for such a patch.
D
Dan Irwin
Now lets say you had one drive with a 16mb buffer and 12ms avg seek
time and one with 8mb buffer and a 10ms avg seek time. Which would be
better in pratical application?
time and one with 8mb buffer and a 10ms avg seek time. Which would be
better in pratical application?
C
CJT
Dan said:Now lets say you had one drive with a 16mb buffer and 12ms avg seek
time and one with 8mb buffer and a 10ms avg seek time. Which would be
better in pratical application?
Whichever was cheaper.
D
Dan Irwin
Would any of this change in the case of a laptop?
C
CJT
Dan said:Would any of this change in the case of a laptop?
Perhaps. Another metric to consider in the case of a laptop is power
consumption. You might be willing to pay a bit of a premium for
something that won't drain your battery.
I
Impmon
Now lets say you had one drive with a 16mb buffer and 12ms avg seek
time and one with 8mb buffer and a 10ms avg seek time. Which would be
better in pratical application?
seek time has little bearing these day. mS is less than a blink of
your eye.
Back in 80's, seek time used to matter a lot as they were much slower
than they are today. Ultimately, I'd choose whichever that costs
less, probably the one with 8MB buffer.
M
Mike Redrobe
Impmon said:seek time has little bearing these day. mS is less than a blink of
your eye.
Going off topic:
CD drives always have slow seek times, but always quote times
excluding spin-up time (and some will get that first access a full
second or two quicker than others)
I`ve never seen a comparative review of CD drives mention this,
which is odd.
F
Folkert Rienstra
Impmon said:seek time has little bearing these day.
Clueless. It has actually much more bearing 'these days'.
mS is less than a blink of your eye.
And 100 blinks is a full second, a small lifetime in data transfer time.
The actual data transfer time of 100 5kB files is 10 ms (at 50MB/s media
to buffer) out of a whopping 1410 ms or 1610 ms (*) of total transfer time.
That is a whopping less than 1% spent in actually transferring the data.
If the 500kB were transferred as a single contiguous file it would have
transfered in ~25ms total, ~60 times faster than the seperate files.
* 4 ms latency added to 10 or 12 ms seek time.
Back in 80's, seek time used to matter a lot as they were much slower
than they are today.
Nonsense.
STR has increased a huge amount more than access time decreased.
The gap between STR and average random access transfer rate has
only widened.