Обсуждение: Re: Re: Future plans for raw devices ?
> > Most think that raw devices are a pain and offer little performance > > improvement and lots of portability and coding problems. > > I don't know how much it is a performance improvement (someone say 10-20%), > but Bruce is probably right, it is a huge work and with dependence on > hardware & system implementation. > > We already discussed about it --- it is hacker's archive. One intresting issue is that commerical databases that recommended raw spaces are moving away from them, which helps us to know that the raw device benefit must be pretty small. > > IMHO now is not in PG background for features like I/O raw or on-line > replication. It needs better storage layout and tablespace feature. Vadim is working on storage layout for 7.2, and replication should be done after WAL is implemented. -- Bruce Momjian | http://www.op.net/~candle pgman@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026
On Fri, 2 Jun 2000, Bruce Momjian wrote:
> > > Most think that raw devices are a pain and offer little performance
> > > improvement and lots of portability and coding problems.
> >
> > I don't know how much it is a performance improvement (someone say 10-20%),
> > but Bruce is probably right, it is a huge work and with dependence on
> > hardware & system implementation.
> >
> > We already discussed about it --- it is hacker's archive.
>
> One intresting issue is that commerical databases that recommended raw
> spaces are moving away from them, which helps us to know that the
> raw device benefit must be pretty small.
Yes. Before one year I thought that I/O raw is very good feature. Hmm, but
if I a little explore something about it, I in current time not sure.
Primaty must be good operating system. A raw is a jink only :-)
> > IMHO now is not in PG background for features like I/O raw or on-line
> > replication. It needs better storage layout and tablespace feature.
>
> Vadim is working on storage layout for 7.2, and replication should be
> done after WAL is implemented.
I know. Will possible create on-line replication via WAL? Probably not,
because it needs locking over more backend. Has this feature (on-line
replication) anywise SQL engine? Sybase?
Karel
On Fri, 2 Jun 2000, Bruce Momjian wrote: > > > Most think that raw devices are a pain and offer little performance > > > improvement and lots of portability and coding problems. > > > > I don't know how much it is a performance improvement (someone say 10-20%), > > but Bruce is probably right, it is a huge work and with dependence on > > hardware & system implementation. > > > > We already discussed about it --- it is hacker's archive. > > One intresting issue is that commerical databases that recommended raw > spaces are moving away from them, which helps us to know that the > raw device benefit must be pretty small. I beg to differ. Informix (the system my shop uses) blazes using raw devices; Oracle (from what I understand) does also. Sybase (a resource hog), on the other hand, seems to prefer cooked file systems. > > > > IMHO now is not in PG background for features like I/O raw or on-line > > replication. It needs better storage layout and tablespace feature. > > Vadim is working on storage layout for 7.2, and replication should be > done after WAL is implemented. > > -- > Bruce Momjian | http://www.op.net/~candle > pgman@candle.pha.pa.us | (610) 853-3000 > + If your life is a hard drive, | 830 Blythe Avenue > + Christ can be your backup. | Drexel Hill, Pennsylvania 19026 > Max Pyziur BRAMA - Gateway Ukraine pyz@brama.com http://www.brama.com/
> On Fri, 2 Jun 2000, Bruce Momjian wrote: > > > > > Most think that raw devices are a pain and offer little performance > > > > improvement and lots of portability and coding problems. > > > > > > I don't know how much it is a performance improvement (someone say > 10-20%), > > > but Bruce is probably right, it is a huge work and with dependence on > > > hardware & system implementation. > > > > > > We already discussed about it --- it is hacker's archive. > > > > One intresting issue is that commerical databases that recommended raw > > spaces are moving away from them, which helps us to know that the > > raw device benefit must be pretty small. > > I beg to differ. Informix (the system my shop uses) blazes using raw > devices; Oracle (from what I understand) does also. Sybase (a resource > hog), on the other hand, seems to prefer cooked file systems. I know about Informix, but I heard Oracle was moving away from raw devices, and the speed improvement was only a few percentage points. -- Bruce Momjian | http://www.op.net/~candle pgman@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026
> > > On Fri, 2 Jun 2000, Bruce Momjian wrote: > > > > > > > Most think that raw devices are a pain and offer little performance > > > > > improvement and lots of portability and coding problems. > > > > > > > > I don't know how much it is a performance improvement (someone say > > 10-20%), > > > > but Bruce is probably right, it is a huge work and with dependence on > > > > hardware & system implementation. > > > > > > > > We already discussed about it --- it is hacker's archive. > > > > > > One intresting issue is that commerical databases that recommended raw > > > spaces are moving away from them, which helps us to know that the > > > raw device benefit must be pretty small. > > > > I beg to differ. Informix (the system my shop uses) blazes using raw > > devices; Oracle (from what I understand) does also. Sybase (a resource > > hog), on the other hand, seems to prefer cooked file systems. > > I know about Informix, but I heard Oracle was moving away from raw > devices, and the speed improvement was only a few percentage points. For Informix it's far more than a few percentage points; and then when you add striping you move even faster. > -- > Bruce Momjian | http://www.op.net/~candle > pgman@candle.pha.pa.us | (610) 853-3000 > + If your life is a hard drive, | 830 Blythe Avenue > + Christ can be your backup. | Drexel Hill, Pennsylvania 19026 > Max Pyziur BRAMA - Gateway Ukraine pyz@brama.com http://www.brama.com/
On Fri, Jun 02, 2000 at 02:26:58PM -0400, pyz@panix.com wrote: > > I know about Informix, but I heard Oracle was moving away from raw > > devices, and the speed improvement was only a few percentage points. > > For Informix it's far more than a few percentage points; and then > when you add striping you move even faster. There is nothing stopping you from using striping with filesystems. The overhead of using filesystems are very small, so the only way to get better performance is from doing better caching. In other words, it only makes sence if using raw devices disable the buffer-cache, and the database cache is better than the OS-cache. It will be very different from OS to OS. -- Ragnar Kjørstad