Re: Using a ramdisk

2006-08-18 Thread Scott Ryan
On Thursday 17 August 2006 13:41, Alex Bramley wrote with regard to - Re: 
Using a ramdisk :
 Hi Bjorn,

 Bjorn Jensen wrote:
  Ramprasad wrote:
  On Wed, 2006-08-09 at 10:27 +0200, Bjorn Jensen wrote:
  Can spamassassin benefit in any way from a ramdisk ?
  The server we have for spamassassin, has 3 gigs of ram, and spamd
  doesn't even use 1 gig of that, so I thought perhaps it would speed
  things up if I could place something on a ramdisk. But this leads to
  the question, does spamassassin do any disk intensive things ?
  I'm running that gocr image scanning as well, could this benefit from
  it, or is it the network lookups that are the slow part in any case ?
  Currently a mail is processed in about 1.5 - 6 seconds
 
  regards,
  Bjorn Jensen
 
  Can you get your MTA to write in the ramdisk while it is queing/scanning
  the mail. That is where you will get most of your speed. But this may
  not be a safe option always.
  Typically using scanners like Mailscanner , you could do the actual Mail
  scanning when the mail is on the ramdisk. That gives you good
  performance benefit. http://www.mailscanner.info/serve/cache/120.html
 
  I appreciate the response on this topic, however I was merely interested
  in the spamassassin aspect of this, and if I could gain a benefit there,
  and I'm not interested in any mailserver options at the moment.
 
  But once again, thanks for taking the time to reply.
 
  Regards,
  Bjorn Jensen

 I realise this thread is a week old now, but I thought you might be
 interested to know that we cut disk access by nearly 60% on a set of six
 mailservers by moving the Bayes and auto-whitelist berkeley databases to
   a ramdisk. I'm not sure whether you will be able to achieve a similar
 speed-up (the databases used to be on a RAID-5 filesystem in our case,
 so small writes such as database updates required reads from all other
 disks in the RAID), but there are some definite benefits to be gained
 from doing this. Of course, if you're using an SQL database as your
 bayes store, this is irrelevant.

 Cheers,
 --alex

Just been thinking about this a little bit...
If I am not mistaken the BayesDB is db3? 
Can it not be compiled against DB4, then you can set the memory cache in a 
DB_CONFIG file. This way you get the performance of a ramdisk, but do not run 
the risk of losing the DB... 

just my $0.02...

-- 
Regards,

Scott Ryan
ISP Systems Development  Integration Specialist
Telkom Internet
-
Good judgement comes with experience. 
Unfortunately, the experience
usually comes from bad judgement.
-


Re: Using a ramdisk

2006-08-17 Thread Alex Bramley

Hi Bjorn,

Bjorn Jensen wrote:

Ramprasad wrote:

On Wed, 2006-08-09 at 10:27 +0200, Bjorn Jensen wrote:
 

Can spamassassin benefit in any way from a ramdisk ?
The server we have for spamassassin, has 3 gigs of ram, and spamd 
doesn't even use 1 gig of that, so I thought perhaps it would speed 
things up if I could place something on a ramdisk. But this leads to 
the question, does spamassassin do any disk intensive things ?
I'm running that gocr image scanning as well, could this benefit from 
it, or is it the network lookups that are the slow part in any case ?

Currently a mail is processed in about 1.5 - 6 seconds

regards,
Bjorn Jensen




Can you get your MTA to write in the ramdisk while it is queing/scanning
the mail. That is where you will get most of your speed. But this may
not be a safe option always.
Typically using scanners like Mailscanner , you could do the actual Mail
scanning when the mail is on the ramdisk. That gives you good
performance benefit. http://www.mailscanner.info/serve/cache/120.html

  
I appreciate the response on this topic, however I was merely interested 
in the spamassassin aspect of this, and if I could gain a benefit there, 
and I'm not interested in any mailserver options at the moment.


But once again, thanks for taking the time to reply.

Regards,
Bjorn Jensen



I realise this thread is a week old now, but I thought you might be 
interested to know that we cut disk access by nearly 60% on a set of six 
mailservers by moving the Bayes and auto-whitelist berkeley databases to 
 a ramdisk. I'm not sure whether you will be able to achieve a similar 
speed-up (the databases used to be on a RAID-5 filesystem in our case, 
so small writes such as database updates required reads from all other 
disks in the RAID), but there are some definite benefits to be gained 
from doing this. Of course, if you're using an SQL database as your 
bayes store, this is irrelevant.


Cheers,
--alex


Re: Using a ramdisk

2006-08-17 Thread John D. Hardin
On Thu, 17 Aug 2006, Alex Bramley wrote:

 I realise this thread is a week old now, but I thought you might
 be interested to know that we cut disk access by nearly 60% on a
 set of six mailservers by moving the Bayes and auto-whitelist
 berkeley databases to a ramdisk.

How do you copy the updated AWL and (if you're autolearning) Bayes
databases to fixed storage so that they aren't lost? Is it as simple
as a set of copy commands in a cron.hourly job?

--
 John Hardin KA7OHZICQ#15735746http://www.impsec.org/~jhardin/
 [EMAIL PROTECTED]FALaholic #11174pgpk -a [EMAIL PROTECTED]
 key: 0xB8732E79 - 2D8C 34F4 6411 F507 136C  AF76 D822 E6E6 B873 2E79
---
  Apparently the Bush/Rove idea of being a fiscal conservative is
  to spend money like there's no tomorrow, run up huge deficits, and
  pray the Rapture happens before the bills come due.
   -- atul666 in Y! SCOX forum
---



Using a ramdisk

2006-08-09 Thread Bjorn Jensen

Can spamassassin benefit in any way from a ramdisk ?
The server we have for spamassassin, has 3 gigs of ram, and spamd 
doesn't even use 1 gig of that, so I thought perhaps it would speed 
things up if I could place something on a ramdisk. But this leads to the 
question, does spamassassin do any disk intensive things ?
I'm running that gocr image scanning as well, could this benefit from 
it, or is it the network lookups that are the slow part in any case ?

Currently a mail is processed in about 1.5 - 6 seconds

regards,
Bjorn Jensen

--

A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting



Re: Using a ramdisk

2006-08-09 Thread jdow

From: Bjorn Jensen [EMAIL PROTECTED]


Can spamassassin benefit in any way from a ramdisk ?
The server we have for spamassassin, has 3 gigs of ram, and spamd 
doesn't even use 1 gig of that, so I thought perhaps it would speed 
things up if I could place something on a ramdisk. But this leads to the 
question, does spamassassin do any disk intensive things ?
I'm running that gocr image scanning as well, could this benefit from 
it, or is it the network lookups that are the slow part in any case ?

Currently a mail is processed in about 1.5 - 6 seconds


Ramdisk is no different than keeping it daemonized in ram except that
you'd still have to go through the spamassassin perl load. That is
why spamc/spamd exist.



--

A: Because it messes up the orderObservation. Side posting has
in which people normally read text.  the best chance of preserving
Q: Why is it such a bad thing?   context, doesn't it?
A: Top-posting


{^,-}


Re: Using a ramdisk

2006-08-09 Thread Daryl C. W. O'Shea

Bjorn Jensen wrote:

Can spamassassin benefit in any way from a ramdisk ?
The server we have for spamassassin, has 3 gigs of ram, and spamd 
doesn't even use 1 gig of that, so I thought perhaps it would speed 
things up if I could place something on a ramdisk. But this leads to the 
question, does spamassassin do any disk intensive things ?
I'm running that gocr image scanning as well, could this benefit from 
it, or is it the network lookups that are the slow part in any case ?

Currently a mail is processed in about 1.5 - 6 seconds


You could stick your /tmp partition on a ramdisk and in theory get a 
little more performance, but most of your 1.5 - 6 seconds is waiting for 
DNS queries to return (that's why it's so variable).


If you're looking for more throughput, rather than a reduction in 
individual scan times, it sounds like you've got the memory for more 
children.



Daryl


Re: Using a ramdisk

2006-08-09 Thread Ramprasad
On Wed, 2006-08-09 at 10:27 +0200, Bjorn Jensen wrote:
 Can spamassassin benefit in any way from a ramdisk ?
 The server we have for spamassassin, has 3 gigs of ram, and spamd 
 doesn't even use 1 gig of that, so I thought perhaps it would speed 
 things up if I could place something on a ramdisk. But this leads to the 
 question, does spamassassin do any disk intensive things ?
 I'm running that gocr image scanning as well, could this benefit from 
 it, or is it the network lookups that are the slow part in any case ?
 Currently a mail is processed in about 1.5 - 6 seconds
 
 regards,
 Bjorn Jensen
 

Can you get your MTA to write in the ramdisk while it is queing/scanning
the mail. That is where you will get most of your speed. But this may
not be a safe option always. 

Typically using scanners like Mailscanner , you could do the actual Mail
scanning when the mail is on the ramdisk. That gives you good
performance benefit. 
http://www.mailscanner.info/serve/cache/120.html

Thanks
Ram







RE: Using a ramdisk

2006-08-09 Thread Xander

When using amavisd-new a ramdisk. Ralf Hildebrandt has setup a small page
about it:

http://www.stahl.bau.tu-bs.de/~hildeb/postfix/amavisd_tmpfs.shtml

He says it gives some optimalization.

Regards,

Xander



From: Bjorn Jensen [EMAIL PROTECTED]

 Can spamassassin benefit in any way from a ramdisk ?
 The server we have for spamassassin, has 3 gigs of ram, and spamd 
 doesn't even use 1 gig of that, so I thought perhaps it would speed 
 things up if I could place something on a ramdisk. But this leads to the 
 question, does spamassassin do any disk intensive things ?
 I'm running that gocr image scanning as well, could this benefit from 
 it, or is it the network lookups that are the slow part in any case ?
 Currently a mail is processed in about 1.5 - 6 seconds

Ramdisk is no different than keeping it daemonized in ram except that
you'd still have to go through the spamassassin perl load. That is
why spamc/spamd exist.

 
 -- 
 
 A: Because it messes up the orderObservation. Side posting has
 in which people normally read text.  the best chance of preserving
 Q: Why is it such a bad thing?   context, doesn't it?
 A: Top-posting

{^,-}



Re: Using a ramdisk

2006-08-09 Thread Bjorn Jensen

Ramprasad wrote:

On Wed, 2006-08-09 at 10:27 +0200, Bjorn Jensen wrote:
  

Can spamassassin benefit in any way from a ramdisk ?
The server we have for spamassassin, has 3 gigs of ram, and spamd 
doesn't even use 1 gig of that, so I thought perhaps it would speed 
things up if I could place something on a ramdisk. But this leads to the 
question, does spamassassin do any disk intensive things ?
I'm running that gocr image scanning as well, could this benefit from 
it, or is it the network lookups that are the slow part in any case ?

Currently a mail is processed in about 1.5 - 6 seconds

regards,
Bjorn Jensen




Can you get your MTA to write in the ramdisk while it is queing/scanning
the mail. That is where you will get most of your speed. But this may
not be a safe option always. 


Typically using scanners like Mailscanner , you could do the actual Mail
scanning when the mail is on the ramdisk. That gives you good
performance benefit. 
http://www.mailscanner.info/serve/cache/120.html


  
I appreciate the response on this topic, however I was merely interested 
in the spamassassin aspect of this, and if I could gain a benefit there, 
and I'm not interested in any mailserver options at the moment.


But once again, thanks for taking the time to reply.

Regards,
Bjorn Jensen

--

A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting



Re: Using a ramdisk

2006-08-09 Thread Bjorn Jensen

Daryl C. W. O'Shea wrote:

Bjorn Jensen wrote:

Can spamassassin benefit in any way from a ramdisk ?
The server we have for spamassassin, has 3 gigs of ram, and spamd 
doesn't even use 1 gig of that, so I thought perhaps it would speed 
things up if I could place something on a ramdisk. But this leads to 
the question, does spamassassin do any disk intensive things ?
I'm running that gocr image scanning as well, could this benefit from 
it, or is it the network lookups that are the slow part in any case ?

Currently a mail is processed in about 1.5 - 6 seconds


You could stick your /tmp partition on a ramdisk and in theory get a 
little more performance, but most of your 1.5 - 6 seconds is waiting 
for DNS queries to return (that's why it's so variable).


If you're looking for more throughput, rather than a reduction in 
individual scan times, it sounds like you've got the memory for more 
children.


I was more hoping I could speed things up for the existing mails, since 
our system is not struggling at the moment, and the mailserver is the 
bottleneck, not the spamassassin server.


But I guess this answers my question, thanks :)

Bjorn

--

A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting