difference between BufferedInputStream(or output) and other i/ostreams,regarding buffers.

[nagarjuna borra]

i want to know the difference between the buffer associated with BufferedInputStream and the bytearray used in read(byte []) of InputStream.

to me both look the same,both are located in random access memory,so how does BufferedInputStream improve performance ??

(I remember usage of buffers in C with read() system call,and read(byte []) of java(InputStream class) looks the same).

[Stas Melnychenko]

InputStream is abstract.

[Jeff Verdegan]

Basically, BOS is guaranteed to buffer when it makes sense, whereas plain old OS makes no such guarantee. It is allowed to simply loop over the array, calling lower-level I/O writes for each byte, incurring, for example, network or file system overhead for each byte. Same thing on the input side.

[Jeff Verdegan]

Stas Melnychenko wrote:InputStream is abstract.

That doesn't really answer the question at all.

And note that, for instance, OutputStream.write(byte b[]) is not abstract.

[Stas Melnychenko]

Jeff Verdegan wrote:

Stas Melnychenko wrote:InputStream is abstract.

That doesn't really answer the question at all.

And note that, for instance, OutputStream.write(byte b[]) is not abstract.

Tnx, I understood.

[nagarjuna borra]

Jeff Verdegan wrote:Basically, BOS is guaranteed to buffer when it makes sense, whereas plain old OS makes no such guarantee. It is allowed to simply loop over the array, calling lower-level I/O writes for each byte, incurring, for example, network or file system overhead for each byte. Same thing on the input side.

I didnt get that,i was thinking lower level I/O is faster..

did you mean lower level I/O is done in case of buffers & overhead is associated with byte arrays??

[Jeff Verdegan]

nagarjuna borra wrote:

Jeff Verdegan wrote:Basically, BOS is guaranteed to buffer when it makes sense, whereas plain old OS makes no such guarantee. It is allowed to simply loop over the array, calling lower-level I/O writes for each byte, incurring, for example, network or file system overhead for each byte. Same thing on the input side.

I didnt get that,i was thinking lower level I/O is faster..

Every time we physically put bytes onto a disk or network, there is an overhead. For instance, if I send 1 single byte of data over a TCP socket, I will be sending at least something like 15 bytes--14 bytes of TCP segment overhead for stuff like source port, dest port, sequence number, etc., plus our 1 data byte. And that doesn't even count overhead for the IP packet, which I think is at least 16 bytes.

So, which would you prefer?

// THIS : 32 * (1 + 32) bytes on the wire, 32 calls to the physical I/O system
low_level_networkwrite(my_single_byte); // my data byte #1, (1 + 32) bytes on the wire
low_level_networkwrite(my_single_byte); // my data byte #2, (1 + 32) bytes on the wire
...
low_level_networkwrite(my_single_byte); // my data byte #32, (1 + 32) bytes on the wire


// OR THIS : 1 * (32 + 32) bytes on the wire, 1 call to the physical I/O system
low_level_networkwrite(my_32_bytes); // my data bytes #1-#32, 32 + 32 bytes on the wire

[nagarjuna borra]

byte to byte&chunks of bytes-got that,thanks.