|
Dec. 2002 To-dos:
1.
λRAM - Benchmark
Experiments;
2. Further studying: DSM, VTK
Interaction, Data Mining, OO Programming;
3. Interaction in VTK;
Mon |
Tue |
Wed |
Thu |
Fri |
Sat |
Sun |
|
|
|
|
|
|
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
10 |
11 |
12 |
13 |
14 |
15 |
16 |
17 |
18 |
19 |
20 |
21 |
22 |
23 |
24 |
25 |
26 |
27 |
28 |
29 |
30 |
31 |
|
|
|
|
|
- Final Week
|
* Three section: 1) implementation of mmap;
2) kiobuf mechanism and 3) DMA I/O operations
* Three level page table in
Linux: PGD (page directory), PMD (page mid-level Directory) and
Page Table - referred in <asm/page.h> and <asm/pgtable.h>
How to map virtual address to physical
memory
|
|
|
1) λRAM borrows some idea from DSM, more
complicated than mmap
2) λRAM is a multi-level architecture
3) Latency is the more fatal to λRAM, since
λRAM fetch thru NIC from remote server
More detail needed; but mmap is a good analog.
Reference - Linux Device Drivers, 2nd Edition
Chapter 13 - mmap and DMA.
|
|
|
1) virtual memory allocation uses malloc(); the
latter - mmap()
2) DSM is just a mechanism which handles the
distributed memory on cluster as unified address space; besides
that, λRAM should prefecth the data from remote server on
the other side in the WAN.
3) no write operation in λRAM. how will
λRAM work if writing is required? - so that it is the almost
the same as DSM which will allocate virtual address space.
To sum up, DSM provides a block of memory
(physical and virtual) for the processes; λRAM generates a
mapping from network transfer pipeline to address space.
λRAM should work like mmap:
mmap - void* mmap(void*, size_t, ....)
LamdaRAM - LamdaRAM* LamdaMap(void*, size_t,
....)
Reference
1. DSM -
2. Computer Architecture - A Quantitative
Approach, Ch8.4
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1. Using TCPServer to listen the Query request
2. Using TCPClient to send query request
3. Using rbudpSender to send data
4. Using rbudpReceive to receive data
|
|
|