Re: [Nbd] RFC: tests too slow?
- To: Wouter Verhelst <w@...112...>
- Cc: nbd-general@lists.sourceforge.net
- Subject: Re: [Nbd] RFC: tests too slow?
- From: Alex Bligh <alex@...872...>
- Date: Thu, 09 Jun 2011 16:28:10 +0100
- Message-id: <95A3B5A25650D6E0B295C08D@...873...>
- Reply-to: Alex Bligh <alex@...872...>
- In-reply-to: <20110609144408.GE6299@...510...>
- References: <72176AE5DE4057A1ED9FA3B0@...873...> <20110609144408.GE6299@...510...>
Wouter,
--On 9 June 2011 16:44:08 +0200 Wouter Verhelst <w@...112...> wrote:
As long as the tests don't start taking hours, we'll be fine. A few
minutes isn't a big issue.
OK, great.
The only other one I was half planning to add was another integrity
test using midsize writes - i.e. many sectors, but not larger than the
normal write size. With the write-ordering stuff turned on (as it
is by default) you get very low queue depths with the 'huge' integrity
tests as there is such a high percentage chance of overlapped writes.
With the tiny writes, you get big queue depths but practically no
overlap, as overlap is statistically unlikely (and even in the
dbench capture I did, it doesn't read stuff it's written before
because it's all in cache, so whilst it's a good performance test
it isn't a fantastic integrity test). This is just a matter of tweaking
maketr to produce some smaller records and perhaps deliberately
introduce a bit of overlap, as there is more chance of servers
disordering reads/writes if they are of different sizes and thus
take different times. No code changes to nbd-tester-client.
--
Alex Bligh
Reply to: