The system we develop change "readpage" by mpage_read,but now the read I/O performance is not as high as in 2.4.
I compare the read path, found that aio is the main difference.
2.6.9
ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
{
struct kiocb kiocb;
ssize_t ret;

init_sync_kiocb(&kiocb, filp);
kiocb.ki_pos = *ppos;
ret = filp->f_op->aio_read(&kiocb, buf, len, kiocb.ki_pos);
if (-EIOCBQUEUED == ret)
ret = wait_on_sync_kiocb(&kiocb);
*ppos = kiocb.ki_pos;
return ret;
}

2.4.18
static ssize_t
nfs_file_read(struct file * file, char * buf, size_t count, loff_t *ppos)
{
struct dentry * dentry = file->f_dentry;
struct inode * inode = dentry->d_inode;
ssize_t result;

dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
dentry->d_parent->d_name.name, dentry->d_name.name,
(unsigned long) count, (unsigned long) *ppos);

result = nfs_revalidate_inode(NFS_SERVER(inode), inode);
if (!result)
result = generic_file_read(file, buf, count, ppos);
return result;
}