Re: [HACKERS] A Better External Sort?

Jeff, Josh,

On 10/3/05 2:16 PM, "Josh Berkus" <josh(at)agliodbs(dot)com> wrote:

> Jeff,
>
>>> Nope, LOTS of testing, at OSDL, GreenPlum and Sun. For comparison, A
>>> Big-Name Proprietary Database doesn't get much more than that either.
>>
>> I find this claim very suspicious. I get single-threaded reads in
>> excess of 1GB/sec with XFS and > 250MB/sec with ext3.
>
> Database reads? Or raw FS reads? It's not the same thing.
>
> Also, we're talking *write speed* here, not read speed.

I think you are both talking past each other here. I'll state what I
*think* each of you are saying:

Josh: single threaded DB writes are limited to 25MB/s

My opinion: Not if they're done better than they are now in PostgreSQL.
PostgreSQL COPY is still CPU limited at 12MB/s on a super fast Opteron. The
combination of WAL and head writes while this is the case is about 50MB/s,
which is far from the limit of the filesystems we test on that routinely
perform at 250MB/s on ext2 writing in sequential 8k blocks.

There is no reason that we couldn't do triple the current COPY speed by
reducing the CPU overhead in parsing and attribute conversion. We've talked
this to death, and implemented much of the code to fix it, but there's much
more to do.

Jeff: Plenty of FS bandwidth to be had on Linux, observed 250MB/s on ext3
and 1,000MB/s on XFS.

Wow - can you provide a link or the results from the XFS test? Is this 8k
blocksize sequential I/O? How many spindles and what controller are you
using? Inquiring minds want to know...

- Luke