From: Honghai Liu (liu789_at_hotmail.com)
Date: Wed Jan 14 2004 - 16:58:14 PST
Reviewer: Honghai Liu
The paper questions one of the basic ideas we take for granted in conversional OS, address space cannot be shared between processes. It may be largely because the hardware (CPU) used to have only 32-bit support and therefore it is too small for applications if they all shared the same space. Many experiments had been done to simulate a single address space OS, however, the complexity and cost were too high to attract real commercial applications.
With the appearance of 64-bit address space CPU, it is reasonable to have Opal, a single address OS, to demonstrate the power of taking the advantage of such a huge space.
Key ideas/takes:
First: If the address space is not context-dependant, sharing become easily: every object references are global and all time (possibly in the networks!). This concept is most attractive because a lot of the problems comes with conversional systems can have a potentially better answer (naming, synchronization, timing etc.)
Second, protection can be separated from addressing. This may not be regarded an obvious advantage. Opal still have segment-grained protection, however, fine-grained protection is left to language and runtime systems.
Third, performance gains are promising for those applications, which require extensive integrations. This was evidenced by analysis of Boeing CAD system; nevertheless, it may be more interesting to quantify the performance in the paper.
By the way, one unexpected and intriguing problem the paper points out is that copying could be a problem and may requires translation.
This archive was generated by hypermail 2.1.6 : Wed Jan 14 2004 - 16:58:18 PST