KR19980030106A - How to send LAN driver message of main computer - Google Patents

Abstract

In the present invention, when a CPU is transmitted from a CPU kernel to an I / O processor kernel, a local BUF is allocated and returned to a remote BUF. When the transmission and reception data are not balanced, one BUF is insufficient. It relates to a LAN driver message transmission method of the host computer to prevent the degradation of performance and stability.

The present invention comprises a local buffer consisting of a plurality of buffers used when transferring data from the CPU to the I / O processor kernel and a remote buffer consisting of a plurality of buffers used when transferring data from the I / O processor kernel to the CPU. In the mutual data transfer method, the buffer allocated from the local buffer is returned to the local buffer from the I / O processor kernel, and the buffer allocated from the remote buffer is returned to the remote buffer from the central processing unit kernel to transmit mutual data.

Description

How to send LAN driver message of main computer

1 is a BPE pool configuration diagram of a conventional main computer.

2 is a schematic diagram of the EPB pool of a conventional main computer.

3 is a BUF pool configuration diagram of a conventional main computer.

4 is a BUF pool configuration diagram of the main computer according to the present invention.

The present invention relates to LAN (LAN) driver message transmission of the host computer, and in particular, when transmitting data from the CPU kernel to the I / O kernel, the local BUF is allocated to the remote BUF and is returned to the remote BUF. The present invention relates to a LAN driver message transmission method of a host computer for preventing performance and stability deterioration caused by insufficient BUF when received data is not balanced.

In general, an area that can be shared by the CPU kernel and the I / O processor kernel is allocated to the CPU kernel area to configure a BPP channel structure.

In the BPP channel structure, data transmission is performed by recording data to be transmitted in a shared area, and the other party reads the information.

The data transfer method is similarly performed between the MVME374 driver of the central processing unit kernel layer, the MVME374 driver of the IOP kernel layer, the MVME374 driver of the input / output processor kernel layer, and the MVME374 controller board.

The shared interface memory pool is composed of BPE, EPB, and BUF. The BPE includes information for handling BPP channels in data transmission, and has address information on EPBs to be transmitted.

The EPB contains the header information necessary for transmission and also contains the address information for the BUF to be transmitted.

The BUF is where the actual data to be transmitted exists.

The BPE, EPB, and BUF pools consist of local and remote pools, respectively. The local pool is the pool area used by the central processing unit kernel area, and the remote pool is the pool area used by the I / O processor kernel area.

1 is a structure of the BPE pool described above.

As shown in this figure, when transferring data from the central processing unit kernel to the I / O processor kernel, the BPE must be allocated. If the BPE is suspended in lc.bpp_write after allocating the data in lc.get_bpe and processing it for transfer, the I / O processor kernel In rep> bpp_read, BPE is taken by sending process.

The I / O processor kernel does the necessary work and returns the BPE to re-> ret_bpe.

Next, FIG. 2 is a structure of the EPB pool described above.

When transferring data from the central processing unit kernel to the I / O processor kernel, an EPB must be allocated. After processing from the lr.epb_head, the processing is required for the transfer and suspended at the BPE allocated above.

The I / O processor kernel does the necessary processing and returns the EPB to rr.epb_tail.

Finally, Figure 3 is the structure of the BUF pool described above.

The BUF is where the actual data to be sent exists.

When data is transferred from the central processing unit kernel to the I / O processor kernel, a BUF is allocated. The data is allocated from lr.buf_head, copied to the BUF, and then suspended in the pre-allocated EPB.

The I / O processor kernel does the necessary work and returns the BUF to rr.buf_taiL.

That is, in the BUF pool structure, the BUF is the part where actual data to be transmitted exists. When data is transferred from the central processing unit kernel to the I / O processor kernel, the data received from lr.buf_head is copied to the BUF, and the monthly data is allocated to the pre-allocated EPB. The I / O processor kernel returns the BUF to rr.buf_tail after performing the necessary processing. At this time, the BUF pool structure consists of the local BUF and the remote BUF, and returns to the remote BUF allocated from the local BUF.

Therefore, this structure is not a problem when the transmission and reception data are balanced, but when it is not (the transmission and reception data is not balanced), one BUF is insufficient, which causes problems in system performance and stability. Occurred.

Therefore, the present invention has been proposed to solve the system performance and stability problems occurring in the conventional BUF pool structure as described above, and an object of the present invention is to transfer data from the CPU kernel to the I / O processor kernel. In this case, the local BUF is allocated from the local BUF and returned to the remote BUF. In this case, when the transmission and reception data are not balanced, it provides a LAN driver message transmission method of the host computer to prevent performance and stability deterioration caused by insufficient BUF. .

A method for achieving the object of the present invention, a local buffer consisting of a plurality of buffers used when transferring data from the CPU kernel to the I / O processor kernel and a plurality of data used to transfer data from the I / O processor kernel to the CPU kernel In the method for transmitting mutual data consisting of a remote buffer consisting of a buffer, the buffer allocated from the local buffer is returned to the local buffer from the input and output processor kernel, and the buffer allocated from the remote buffer from the central processing unit kernel to the remote buffer Returning data to each other.

4 is a structure of a BUF pool according to the present invention.

This shows the structure of the new BUF pool by changing lr.buf_tail to rr.buf_tail and rr.buf_tail to lr.buf_tail in the structure of FIG. 3 which is a conventional BUF pool structure.

As shown in this figure, when data is transmitted from the CPU to the I / O processor kernel, a BUF is allocated. The data is allocated from lr.buf_head, copied to the BUF, and then suspended in the pre-assigned EPB.

The I / O processor kernel does the necessary work and returns the BUF to rr.buf_tail.

When data is transferred from the I / O processor kernel to the central processing unit kernel, a BUF is allocated. The data received from rr.buf_head is copied to the BUF and then suspended in the pre-allocated EPB.

The central processing unit kernel does the necessary processing and returns the BUF to rr.buF_tail.

That is, the BUF allocated from the local BUF is returned to the local BUF from the I / O kernel, and the BUF allocated from the remote BUF is returned from the central processing unit kernel to the remote kernel.

As described above, in the present invention, when the BUF allocated from the local BUF is returned to the local BUF from the input / output kernel, and the BUF allocated from the remote BUF is returned from the central processing unit kernel to the remote kernel, the transmission and reception data are not balanced. However, there is no shortage of one BUF, which contributes significantly to the performance and stability of the system.

Claims (1)

Mutual data transfer consists of a local buffer consisting of a number of buffers used to transfer data from the central processing unit kernel to the I / O processor kernel, and a remote buffer consisting of a plurality of buffers used to transfer data from the I / O processor kernel to the central processing unit kernel. The method of claim 1, wherein the buffer allocated by the local buffer is returned to the local buffer by the input / output processor kernel, and the buffer allocated by the remote buffer is returned to the remote buffer by the central processing unit kernel to transfer mutual data. How to send LAN driver message.