| 1 |
Frank Peters wrote: |
| 2 |
> On Tue, 19 May 2009 15:18:13 +0200 |
| 3 |
> Branko Badrljica <brankob@××××××××××.com> wrote: |
| 4 |
> |
| 5 |
>> I have /var/tmp on ext4 with delayed allocation on ext4 and 8GB RAM. |
| 6 |
>> For me tmpfs wasn't worth the hassle, so i dropped it. I couldn't tell |
| 7 |
>> the difference. |
| 8 |
> |
| 9 |
> In all this discussion about tmpfs, there has been no mention of the |
| 10 |
> gcc "-pipe" option, which instructs the compiler to use RAM instead |
| 11 |
> creating temporary files. Since gcc is the workhorse of the emerge |
| 12 |
> process, it would seem that the "-pipe" option should accomplish |
| 13 |
> results close to that observed with tmpfs. |
| 14 |
|
| 15 |
Well, not even close, unfortunately. -pipe just passes the assembly |
| 16 |
source to the assembler by pipe, nothing else. That's like just 5% of |
| 17 |
the total I/O involved. The 95% of the rest (reading the source from |
| 18 |
disk, writing object code, executables, installing it in a faked root |
| 19 |
environment) is not covered by -pipe. And don't forget the minimized |
| 20 |
disk fragmentation too since all happens in RAM before merging for real. |
Archives