Messages in this thread | | | Date | Sat, 02 Sep 2000 21:31:10 -0600 | From | "Jeff V. Merkey" <> | Subject | Re: zero-copy TCP |
| |
Petr,
I could give you a packet burst module to study. It's MANOS version and not linux, so you will need to backport it, but it will let you fake out a NetWare server without needing all this memory.
Jeff
Petr Vandrovec wrote: > > On Sun, Sep 03, 2000 at 12:00:13AM +0100, Alan Cox wrote: > > > **NCP** is does not reside in IPX at all. That's why the implementation > > > in Linux is busted. The window size is variable and keys off of HOW > > > MANY FREE ECBS ARE IN THE SYSTEM. It doe not belong in the IPX/SPX > > > stack, but inside of MARS-NWE proper. > > > > Packet burst is in MARS-NWE not the kernel IPX code. Our IPX code actually > > doesnt know about NCP at all - so it has no fast paths or kernel helpers for > > the cache. It means we are materially poorer but we dont have to maintain > > magic for NCP. If you were doing a Linux netware clone you would want a > > kncpd for the cache hit cases > > Hi Alan, > there is ncpfs. It is client, not server... Main problem is that > packet burst protocol is that it is only for file reads and writes, > you cannot use it to listing (large) directories, or to return large > replies from NDS in one exchange. > > Fortunately NW5 can do NCP over TCP and UDP and ncpfs (in 2.4) currently > uses 60KB NCP window when talking over TCP to NW5. Due to nature of NCP > protocol it is much faster than with 1KB (or 1.5KB) window used for IPX or > UDP... Unfortunately, server allocates this handshaked window size of memory > for your connection, so 1000 conns with 60KB handshaked window consume 60MB > of memory... > > I do not plan implementing packet burst in ncpfs at all... NW5 does IP > reasonably well to let IPX die. > > > > > The I/O size was typically 512-1024 bytes so you couldnt do page flipping, but > > > > since these were all trusted paths nobody I suspect worried to much. > > > > > > 576 bytes. Drew came up with this because he did not want to fuck > > > > (minus headers) > > Fortunately when Novell included packetburst code, they also added real > IPX MTU discovery into NCP (unfortunately, they really build it into NCP, > so f.e. SPX2 uses different handshake). That way it usually uses 1.45KB > window on ethernet and 4.2kB on tokenring (unless you changed some > settable value in server configuration - then you can go up to your > 16KB tokenring limit). > Best regards, > Petr Vandrovec > vandrove@vc.cvut.cz - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org Please read the FAQ at http://www.tux.org/lkml/
| |