Re: TMPDIR as a tmpfs

2010-06-26 Thread Daryl C. W. O'Shea
On 22/06/2010 10:52 AM, Henrique Fernandes wrote:
 It is safe to use spamassassin tmpdir on a tmpfs mounted system ?

Yes it's safe.

 And if its safe it would have a better performance ?

Potentially.  If you've got memory free for it, it certainly shouldn't
perform worse.

Daryl



Re: TMPDIR as a tmpfs

2010-06-26 Thread jdow

From: Daryl C. W. O'Shea spamassas...@dostech.ca
Sent: Saturday, 2010/June/26 15:23



On 22/06/2010 10:52 AM, Henrique Fernandes wrote:

It is safe to use spamassassin tmpdir on a tmpfs mounted system ?


Yes it's safe.


And if its safe it would have a better performance ?


Potentially.  If you've got memory free for it, it certainly shouldn't
perform worse.


That might be a big if with a huge downside, Daryl.

If the memory used by tmpfs forces SpamAssassin into memory swapping
any speed advantages are more than merely wiped out, aren't they?

{o.o}- her number 1 SA rule is to avoid swap at all costs.


Re: TMPDIR as a tmpfs

2010-06-26 Thread RW
On Sat, 26 Jun 2010 16:33:32 -0700
jdow j...@earthlink.net wrote:


  Potentially.  If you've got memory free for it, it certainly
  shouldn't perform worse.
 
 That might be a big if with a huge downside, Daryl.
 
 If the memory used by tmpfs forces SpamAssassin into memory swapping
 any speed advantages are more than merely wiped out, aren't they?

No, because, at worst, you are exchanging one type of disk access for
another.

A tmpfs partition is memory-backed by swap, a file on a normal
filesystem is cached in memory. There's not all that much difference,
they are both memory backed by a physical disk backing-store. A well
designed kernel will place the physical memory where it's most
effective, whether that's caching a file or keeping the tmpfs or
process pages in ram.

The advantage of tmpfs is not that it's stored in memory, it's that the
kernel can put-off updating the backing store indefinitely - a
temporary file can be created, updated and deleted without troubling
the hard drive.


Re: TMPDIR as a tmpfs

2010-06-23 Thread Matus UHLAR - fantomas
On 22.06.10 12:40, Henrique Fernandes wrote:
 My ram dos not get full, i do not have so many process, i limit it in
 postfix.
 
 It reduces the chances of losing emails if i do not have many process of
 spamassassin runing.
 
 So is safe or not to use tmpfs for tempdir in spamassassin. ?

I use tmpfs for /tmp for years. I don't remember any problems with it.
And it increases system performance in many cases.
Note that /tmp is for TEMPORARY storage, that doesn't need to retain after
reboot.

 This way, everything that spamassassin have to do with the message it does
 on tmpfs.

does SA use /tmp at all?
-- 
Matus UHLAR - fantomas, uh...@fantomas.sk ; http://www.fantomas.sk/
Warning: I wish NOT to receive e-mail advertising to this address.
Varovanie: na tuto adresu chcem NEDOSTAVAT akukolvek reklamnu postu.
99 percent of lawyers give the rest a bad name. 


Re: TMPDIR as a tmpfs

2010-06-23 Thread RW
On Tue, 22 Jun 2010 19:47:00 -0700
Gary Smith gary.sm...@holdstead.com wrote:

  I had a very dramatic performance improvement by switching bayes and
  awl
  databases to MySQL instead of the default BerkeleyDB.  It costs more
  RAM, CPU, and disk space, but scan times reduced dramatically.  I'm
  certain we were I/O bound before this change because we had plenty
  of RAM and CPU available.
  
 
 I agree to the bayes DB being MySQL.  When we switched to that years
 ago it was night and day. 

 

It'd be interesting to see the effect of bayes_learn_to_journal on gdb.
No-one ever mentions trying it, but it might be faster than MySQL.

AFAIK there's no equivalent for AWL, so that would have to be
turned-off to see the benefit.





Re: TMPDIR as a tmpfs

2010-06-22 Thread Eduardo Casarero
2010/6/22 Henrique Fernandes sf.ri...@gmail.com

 It is safe to use spamassassin tmpdir on a tmpfs mounted system ?

 And if its safe it would have a better performance ?

 Here where i work we have big problems with the hard drives, because we
 basically are sharing virtual machines disk over nfs. and spamassasin is a
 virtual machine.

 Any other tips for better performance ?




 []'sf.rique



RE: TMPDIR as a tmpfs

2010-06-22 Thread Gary Smith
 It is safe to use spamassassin tmpdir on a tmpfs mounted system ?
 And if its safe it would have a better performance ?
 Here where i work we have big problems with the hard drives, because we 
 basically are sharing virtual machines disk over nfs. and spamassasin is a 
 virtual machine.
 Any other tips for better performance ?

Ram, lots of it.  This seem to have the biggest impact on my systems.  The only 
time I see disk thrashing on our SA and ClamAV VM's is when they are starved 
for ram.


Re: TMPDIR as a tmpfs

2010-06-22 Thread Henrique Fernandes
My ram dos not get full, i do not have so many process, i limit it in
postfix.

It reduces the chances of losing emails if i do not have many process of
spamassassin runing.

So is safe or not to use tmpfs for tempdir in spamassassin. ?

This way, everything that spamassassin have to do with the message it does
on tmpfs.

[]'sf.rique


On Tue, Jun 22, 2010 at 12:28 PM, Gary Smith gary.sm...@holdstead.comwrote:

  It is safe to use spamassassin tmpdir on a tmpfs mounted system ?
  And if its safe it would have a better performance ?
  Here where i work we have big problems with the hard drives, because we
 basically are sharing virtual machines disk over nfs. and spamassasin is a
 virtual machine.
  Any other tips for better performance ?

 Ram, lots of it.  This seem to have the biggest impact on my systems.  The
 only time I see disk thrashing on our SA and ClamAV VM's is when they are
 starved for ram.



RE: TMPDIR as a tmpfs

2010-06-22 Thread Gary Smith


My ram dos not get full, i do not have so many process, i limit it in postfix.

It reduces the chances of losing emails if i do not have many process of 
spamassassin runing.

So is safe or not to use tmpfs for tempdir in spamassassin. ?

This way, everything that spamassassin have to do with the message it does on 
tmpfs.

---

Outside smtp - postfix - postfix after queue - spamassassin - postfix - 
destination

Even if SA is going slow, no email will be lost.  If it is, something else is 
broken.  If postfix and SA are on the same VM, it's the postfix queue that 
could be slowing things down.  In this case there isn't much you can do as you 
need postfix to be on persistent media.

Also, my understanding is that SA only uses temp files for Razor and DCC 
checks.  Otherwise it should be in ram anyway.  Are you doing Razor or DCC?


Re: TMPDIR as a tmpfs

2010-06-22 Thread Henrique Fernandes
pyzor and dcc

Só it might be it ?

If i am not using wont get in tmp ?


Right now i have 2 server with spamassassin, one i just put everything  in
spamassassin in tmpfs and the other one i did not change anything

[]'sf.rique


On Tue, Jun 22, 2010 at 12:46 PM, Gary Smith gary.sm...@holdstead.comwrote:



 My ram dos not get full, i do not have so many process, i limit it in
 postfix.

 It reduces the chances of losing emails if i do not have many process of
 spamassassin runing.

 So is safe or not to use tmpfs for tempdir in spamassassin. ?

 This way, everything that spamassassin have to do with the message it does
 on tmpfs.

 ---

 Outside smtp - postfix - postfix after queue - spamassassin - postfix
 - destination

 Even if SA is going slow, no email will be lost.  If it is, something else
 is broken.  If postfix and SA are on the same VM, it's the postfix queue
 that could be slowing things down.  In this case there isn't much you can do
 as you need postfix to be on persistent media.

 Also, my understanding is that SA only uses temp files for Razor and DCC
 checks.  Otherwise it should be in ram anyway.  Are you doing Razor or DCC?



Re: TMPDIR as a tmpfs

2010-06-22 Thread Adam Moffett
I don't know if it is safe.  I suspect it will function normally, but I 
think you'd be in danger of losing a few messages on an unexpected reboot.


I had a very dramatic performance improvement by switching bayes and awl 
databases to MySQL instead of the default BerkeleyDB.  It costs more 
RAM, CPU, and disk space, but scan times reduced dramatically.  I'm 
certain we were I/O bound before this change because we had plenty of 
RAM and CPU available.




It is safe to use spamassassin tmpdir on a tmpfs mounted system ?

And if its safe it would have a better performance ?

Here where i work we have big problems with the hard drives, because 
we basically are sharing virtual machines disk over nfs. and 
spamassasin is a virtual machine.


Any other tips for better performance ?




[]'sf.rique




RE: TMPDIR as a tmpfs

2010-06-22 Thread Gary Smith
 I don't know if it is safe.  I suspect it will function normally, but I
 think you'd be in danger of losing a few messages on an unexpected
 reboot.
 
 I had a very dramatic performance improvement by switching bayes and
 awl
 databases to MySQL instead of the default BerkeleyDB.  It costs more
 RAM, CPU, and disk space, but scan times reduced dramatically.  I'm
 certain we were I/O bound before this change because we had plenty of
 RAM and CPU available.
 

I agree to the bayes DB being MySQL.  When we switched to that years ago it was 
night and day.  We have a central MySQL cluster feeding multiple SA instances 
without any problem.  Generally we are running VM's for SA as we can randomly 
spin then up when we need them on machines with idle CPU's.

Gary


Re: TMPDIR as a tmpfs

2010-06-22 Thread Henrique Fernandes
it is taking about 3 seconds each email, i have pyzor and dcc, i am already
runing with mysql db.

After i get the statistis i will post here if tmpfs is faster or not! I made
some script that  can't see anyway for losing email!

Thanks for all advise!

And sorry about my english

[]'sf.rique


On Tue, Jun 22, 2010 at 11:47 PM, Gary Smith gary.sm...@holdstead.comwrote:

  I don't know if it is safe.  I suspect it will function normally, but I
  think you'd be in danger of losing a few messages on an unexpected
  reboot.
 
  I had a very dramatic performance improvement by switching bayes and
  awl
  databases to MySQL instead of the default BerkeleyDB.  It costs more
  RAM, CPU, and disk space, but scan times reduced dramatically.  I'm
  certain we were I/O bound before this change because we had plenty of
  RAM and CPU available.
 

 I agree to the bayes DB being MySQL.  When we switched to that years ago it
 was night and day.  We have a central MySQL cluster feeding multiple SA
 instances without any problem.  Generally we are running VM's for SA as we
 can randomly spin then up when we need them on machines with idle CPU's.

 Gary