Jeff <threshar@torgo.978.org> writes:
> On Tue, 03 Feb 2004 11:46:05 -0500
> Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> I did some idle, very unscientific tests the other day that indicated
>> that MySQL insert performance starts to suck with just 2 concurrent
>> inserters. Given a file containing 10000 INSERT commands, a single
>> mysql client ran the file in about a second. So if I feed the file
>> simultaneously to two mysqls in two shell windows, it should take
>> about two seconds total to do the 20000 inserts, right? The observed
>> times were 13 to 15 seconds. (I believe this is with a MyISAM table,
>> since I just said CREATE TABLE without any options.)
> MyISAM is well known to suck if you update/insert/delete because it
> simply aquires a full table lock when you perform those operations!
Sure, I wasn't expecting it to actually overlap any operations. (If you
try the same test with Postgres, the scaling factor is a little better
than linear because we do get some overlap.) But that shouldn't result
in a factor-of-seven slowdown. There's something badly wrong with their
low-level locking algorithms I think.
regards, tom lane