LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

OLS 2002: Wed

To: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Subject: OLS 2002: Wed
From: "mack@xxxxxxxxxxxx" <mack@xxxxxxxxxxxxxxxxxxxx>
Date: Thu, 27 Jun 2002 12:36:32 -0400 (EDT)
I was wrong about the age of the people doing amazing things with Linux.
Marchello (from Brazil), the person who's taken over producing the 2.4.x
kernels from Linus, is 19.

Last night I left Ratz and Marchello at some pub. Ratz was helping 
Marchello with the 2.4.19 kernel. Ratz was working his way through to the 
bottom of his personal pitcher of beer. I could see from the look in 
Ratz's eyes that if it was humanly possible to get 2.4.19 out last night, 
he was goingo to do it. To help I took his laptop and backpack with me, so 
that he wouldn't be overloaded with unneccessary responsibilities on his 
way home to the jail. I also checked that Ratz had the key code written in 
ball-point pen on his hand. Ratz got back at 4pm, apparently having 
recovered from his preconference 4 day vaction.

One of this morning's talks was an undate by Jeff Dike on user mode Linux
(UML). The amazing this was that Linux could run a copy of itself as a
job, without this being one of the original design criteria. The problem
is that it doesn't do it well. For instance all of the UML processes show
up in the host kernel process table. Clearly only one process, the UML
should show up in the host ps table, while the UML processes whould show
up in the UML process table. Another problem is cooperative memory
management. With only one kernel running, there is no need to swap out
memory, till it is needed. You may as well leave it data in memory for a
few weeks, in case it is ever needed, saving you an IO. With the host 
kernel and the UML kernels all taking all memory possible, there is going 
to be collisions for memory. The UML processes need to take minimum memory 
and the host needs someway of issuing an ultimatum to the UMLs that it 
will be needing memory soon and to either release memory or have it 
swapped out by the host kernel.

Another point from the AMD hammer talk yesterday: with the faster 
hyperlink hardware, the difference in access time for memory on another 
CPU has been reduced from a factor of 10:1 to about 1.5:1, meaning that 
the memory attached to each CPU (your 4G of RAM) can be treated as shared 
memory.

I sat next to Adam Richter of Ygddrasil last night at the reception/film
about Linus/Linux. Ygddrasil only exists now to service requests from
previous customers. Previously it was a viable company employing 8 people.
He regrets that he didn't start looking for venture capital when it was
available. However he realises said that this could have set him up to 
fail in a larger and more spectacular way. I mentioned that it wasn't so 
long ago that the nerds were in charge of computing, that it needed 
someone techinally competent to set up a secure shopping cart, handle 
credit card data securely, track orders etc. Now that is all canned for 
you. He said we were all naive to think that the big companies who were 
handling retailing, wouldn't be able to handle on line ordering or that 
customers who were familiar with the large companies would go elsewhere to 
order on line. So the nerds and geeks had it for about 2yrs in my mind, 
now we're back to the big guys owning it all again.

Joe

-- 
Joseph Mack, mack@xxxxxxxxxxxx
Linux Virtual Server project
http://www.linuxvirtualserver.org



<Prev in Thread] Current Thread [Next in Thread>