T.R | Title | User | Personal Name | Date | Lines |
---|
889.1 | Ingres server I/O is asynchronous | BEAGLE::GODFRIND | TSCG, Valbonne VBO1-2/B15 DTN 828-5163 | Fri Mar 15 1991 18:29 | 12 |
| I take it you mean Ingres on VMS ?
I know that Ingres on Ultrix has to resort to a trick to get asynchronous
writes (because the O/S does not provide them) - it uses a set of child
processes (slaves) to do the I/O, thus freeing the DBMS servers to do something
else while the I/O proceeds.
If I am not mistaken, the servers on VMS also do asynchronous I/O. If you think
about it, how could they _not_ do so ? What would be the point of having
multithreaded servers if all I/O was single-threaded ?
/albert
|
889.2 | Theory vs practice... | BROKE::HIGGS | SQL is a camel in disguise | Fri Mar 15 1991 20:36 | 9 |
| Isn't the point that Rdb/VMS with ACMS should be able to outperform INGRES with
ACMS, in practical situations?
Whether a database uses synchronous or asynchronous writes is simply a technique.
What counts is results, surely?
Or does the customer need theoretical arguments to be convinced?
Bryan
|
889.3 | Ingres logfile writes must be synchronous | BIGUN::PHILLIPS | Blair Phillips, SI, Canberra, Oz | Sun Mar 17 1991 10:12 | 9 |
| Even if Ingres does async i/o elsewhere, it still needs to do
synchronous writes to its logfile when committing a transaction.
Ingres uses a single logfile for all databases on the system, so this
file is a real hot spot in a high update environment. (And no, you
can't put it on a RAM disk, it is essential to database integrity.
ESE20 would be a good move, though.)
Blair
|
889.4 | but what about ... | WARNUT::BRYAN | | Mon Mar 18 1991 12:00 | 6 |
| As an aside, what about the lack of row level locking and the fixed
page size, don't these detract from using Ingres in a TP environment.
Not to mention performance ownership problems.
What do they gain by using the Ingres db ?
|