29728 - nsFileTransport::Process consumes 2 secs out of 19 secs startup

Status: VERIFIED FIXED

(Core :: Networking, defect, P3)

Description

[Suresh Duddi (gone)]

Function				%	#call	millisecs
-----------------------------------------------------------------
nsFileTransport::Process		100.00	 87	2068.45
 nsLocalFileSystem::Open		 49.55	 87	2068.45
 nsPipe::nsPipeOutputStream::WriteFrom	 25.10	154	1047.67
 nsLocalFileSystem::GetInputStream	 20.81	 87	 868.64

We are reading in 87 files on startup. Most of them would be chrome files or 
gif files that dont need a progress (maybe). So just avoiding stat on them 
would save us a bunch. Here is timing on nsLocalFileSystem::Open()

Function				%	#call	millisecs
-----------------------------------------------------------------
do_GetService(nsMIMEService)		51.26	87	1060.35
nsLocalFile::GetFileSize		48.55	87	1004.24

So we do 87 calls to GetFileSize which calls stat. I think the do_GetService() 
is mostly spending time on loading the dll (that is 1 sec wow).

Comment 1

[Warren Harris]

This is mine...

Comment 2

[Suresh Duddi (gone)]

Rehooking into the right dependency for startup performance.

Comment 3

[Warren Harris]

Another thing that nsFileTransport::Process does that is sub-optimal, is every 
time though it's look it acquires and releases it's lock. This lock is used to 
coordinate with request cancelation. This is very friendly to the canceler, but 
takes a lot more time.

The socket transport, on the other hand, only gives up its lock when it needs 
to suspend or resume (because the buffer is full or empty). We should do that 
in the file transport too.

Comment 4

[Warren Harris]

Moving non-essential, non-beta2 and performance-related bugs to M17.

Comment 5

[Warren Harris]

post-beta2

Comment 6

[Warren Harris]

The issue of the lock in the file transport has been addressed. It was removed a 
week or so ago. 

Eliminating GetFileSize (or making it cheaper by consolidating it with other 
stat info) should still be looked at. See bug 18048.

Comment 7

[Warren Harris]

P.S. GetFileSize is related to bug 42772 - eliminate StreamIO::Open

Comment 8

[Warren Harris]

Closing this ancient bug. I've long since removed the lock in 
nsFileTransport::Process, and I'm not sure what else we can do with the 
GetFileSize issue.

Comment 9

[Tom Everingham]

verified