Tuning the performance of ObjectInputStream.readObject()

Question

I'm using JWorks, which roughly corresponds to jdk1.1.8, on my client (VxWorks) and jdk1.4.2 on the server side (Windows, Tomcat).
I'm serializing a big vector and compressing it using GZipOutputStream on the server side and doing the reverse on the client side to recreate the objects.
Server side:
Vector v = new Vector(50000);//also filled with 50k different MyObject instances
ObjectOutputStream oos = new ObjectOutputStream(new GZipOutputStream(socket.getOutputStream()));
oos.writeObject(v);Client side:
ObjectInputStream ois = new ObjectInputStream(new GZipInputStream(socket.getInputStream()));
Vector v = (Vector)ois.readObject();ObjectInputStream.readObject() at the client takes a long time (50secs) to complete, which is understandable as the client is a PIII-700MHz and the uncompressed vector is around 10MB. Now the problem is that because my Java thread runs with real time priority (which is the default on Vxworks) and deprive other threads, including non-Java ones, for this whole 50 sec period. This causes a watchdog thread to reset the system. I guess most of this delay is in GZipInputStream, as part of the un-gzipping process.
My question is, is there a way to make ObjectInputStream.readObject() or any of the other connected streams (gzip and socket) yield the CPU to other threads once in a while? Or is there any other way to deal with the situation?
Is the following a good solution?
class MyGZipInputStream extends GZipInputStream {
    public int count = 0;
    public int read(byte b[], int off, int len) throws IOException {
         if(++count %10 == 0) // to avoid too many yields
              Thread.yield();
          return super.read(b, off,len);
}Thanks
--kannan                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                           

Answer 1

I'd be inclined to put the yielding input stream as close to the incoming data as possible - thus avoiding any risk that time taken to read in data and buffer it will cause the watchdog to trip.I could do that. But as I'm doing Thread.yield only once every x times, would it help much?
Also, as I've now found out, Thread.yield() wouldn't give other low priority tasks a chance to run. So I've switched to Thread.sleep(100) now, even though it could mean a performance hit.
Another relaed question - MyGzipStream.read(byte[], int, int) is called about 3million times during the readObject() of my 10MB vector. This would mean that ObjectInputStream is using very small buffer size. Is there a way to increase it, other than overriding read() to call super.read() with a bigger buffer and then doing a System.arrayCopy()?
Thanks
--kannan