Re: Table Partitioning in Postgres:

From: Jonathan Bartlett <johnnyb(at)eskimo(dot)com>
To: Greg Copeland <greg(at)CopelandConsulting(dot)Net>
Cc: "Shridhar Daithankar<shridhar_daithankar(at)persistent(dot)co(dot)in>" <shridhar_daithankar(at)persistent(dot)co(dot)in>, "PGSQL General (E-mail)" <pgsql-general(at)postgresql(dot)org>
Subject: Re: Table Partitioning in Postgres:
Date: 2003-02-19 16:31:30
Message-ID: Pine.GSU.4.44.0302190827570.14244-100000@eskimo.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-general

> While your statement is correct I did want to clarify that IDE and SCSI
> were lumped together and they should not be. SCSI and IDE performance
> expectations differ because their bus technologies are dramatically
> different. IDE has some serious performance issues with multiple disks
> per channel. A single disk can effectively tie up an IDE channel for

Actually, IDE has performance issues even with only 1 disk per channel.
The SCSI command set allows disconnected operations, so that the kernel
can send commands (get block xxx, get block yyy, get block zzz) and keep
sending commands while the drive processes the answers. With IDE, it is a
synchronous transmission - get blox xxx, wait until disk processes,
receive block xxx, get block yyy, wait until disk processes, receive block
yyy. SCSI disks can also reorder the requests and service them based on
how quickly it can get to each one.

Even on one-channel-per-disk, SCSI wins out.

Jon

> the duration of an outstanding I/O operation, unlike what you would
> expect with SCSI. As such, it is highly recommended to have as many
> channels as disks where you expect concurrent disk I/O on said disks.
>
> In the case of IDE, you should see significant boosts in performance by
> following this tip. Even when sharing a bus with something as simple as
> a CDROM, it can significantly and negatively impact IDE bus performance;
> reflective of both throughput and latency. As such, I highly recommend
> that you use a channel per disk where you expect high rates of
> concurrent disk I/O. As a rule of thumb, you should use a channel per
> disk with IDE where throughput and latency are of concern.
>
>
> Regards,
>
> Greg
>
>
> >
> > >
> > >
> > > > Even Postgresql has to be told to perform vaccum and analyze.
> > > > If the OS had enough intelligence we could trust it to do a good job,
> > > > but until then ...
> > >
> > > Partially true. Postgresql could have done vacuum at runtime at the cost od
> > > performance. So developers delegated the task to admin.
> > >
> > > Looking for a solution in problem, the real benefits won't be visible unless
> > > you put it on a different disk channel. Otherwise RAID is your best bait now
> > > as OS can handle it intelligently and it enhances the IO bandwidth immensely.
> > >
> > > Other than that you can not do much with postgresql right now.
> > >
> > > Shridhar
> > >
> > > ---------------------------(end of broadcast)---------------------------
> > > TIP 1: subscribe and unsubscribe commands go to majordomo(at)postgresql(dot)org
> > >
> >
> >
> > ---------------------------(end of broadcast)---------------------------
> > TIP 5: Have you checked our extensive FAQ?
> >
> > http://www.postgresql.org/users-lounge/docs/faq.html
> --
> Greg Copeland <greg(at)copelandconsulting(dot)net>
> Copeland Computer Consulting
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 1: subscribe and unsubscribe commands go to majordomo(at)postgresql(dot)org
>

In response to

Responses

Browse pgsql-general by date

  From Date Subject
Next Message jerome 2003-02-19 16:35:37 SQL query...
Previous Message Steve Crawford 2003-02-19 16:31:21 Re: continuous data from stdin