X compression techniques (was Re: VNC server based on kdrive using damage extension?)

[Jim Gettys]

On Fri, 2004-02-20 at 15:40, Kurt Pfeifle wrote:
> Jim Gettys wrote:
> 
> > Yes, NX looks very interesting.
> > 
> > SSH by itself makes a big difference, and I understand that NX goes
> > well beyond that. I'll try to somehow dig up the time somewhere.
> > Our paper shows that there is a tremendous lot to be gained
> > by various bandwidth and latency fixes.
> > 
> > The performance issue will primarily come to the fore in environments
> > like LTSP, where scaling on servers is an issue (how many people
> > you can support on a single box).
> 
> I've seen at one time 107 concurrent user sessions on one box (dual
> 2.6 GHz P1V CPU, 4 GByte of RAM), where each user ran a KDE desktop
> with KMail, Konqueror and OpenOffice open, and yet they all had still
> quite responsive systems. IIRC it was a 10 MBit wire (where only a
> smaller part of the users were on switches, the rest on hubs) and the
> most used segment of the net showed only about 40-50% saturation.
> 

Good to hear.

> It looks like a bandwidth of 20 - 40 kBit/sec, RAM of ~40 MByte and
> CPU of 50 MHz per user is enough to easily serve a 100-head group of
> users.
> 

Do we know what the tradeoff is?  Sending less data saves cycles
as well, the equation is complex.

> > NX, or SSH, for an individual
> > user talking to a single machine I'm sure presents little issue.
> 
> Oh, it is. At least for me. My company sends me sometimes to customers
> around the country, to support them onsite. My evenings in the hotel
> rooms have become much more pleasant since the time that I can simply
> use an ISDN dial-in line to access my workstation in our main office.
> At one time, I was even connecting through a 9600 Baud GSM modem link
> and it still somehow worked (slightly better than VNC does inside our
                          - Jim

-- 
Jim Gettys <Jim.Gettys@hp.com>
HP Labs, Cambridge Research Laboratory