You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@trafficserver.apache.org by "Adam W. Dace" <co...@gmail.com> on 2013/10/18 18:01:52 UTC

ATS Storage Question

I've already noticed ATS tends to slow down a bit when using database files
above 2GB...but it got me thinking.

Does anyone know which would be more efficient: 2 database files at 2GB
each...or 4 database files at 1GB each?

Thanks In Advance,

-- 
____________________________________________________________
Adam W. Dace <co...@gmail.com>

Phone: (815) 355-5848
Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>

Google Profile: https://plus.google.com/u/0/109309036874332290399/about

Re: ATS Storage Question

Posted by "Adam W. Dace" <co...@gmail.com>.
No big deal, I was just a bit grumpy that there's no error given.  Happy to
see the documentation is up-to-date.

I think sitting around and simply tuning ATS has made me a bit of a grump
lately.

However, what's really amazing is the speed one can achieve once you get
all the bottlenecks out of the way.
I'm currently tweaking child --> parent failover settings and am kind of in
awe of the speed of this thing.


On Sat, Oct 19, 2013 at 6:16 PM, Alan M. Carroll <
amc@network-geographics.com> wrote:

> Friday, October 18, 2013, 11:30:39 AM, you wrote:
>
> > Even worse, the Admin Guide indicates more than one database file isn't
> currently supported.  Ouch.
>
> That's a bit misleading, it's only one file per physical device. I see
> where that's done but I haven't had time to fix it yet. AFAICT there's no
> real reason for that limitation, it's just an unhappy side effect of how
> the cache spans are calculated.
>
>


-- 
____________________________________________________________
Adam W. Dace <co...@gmail.com>

Phone: (815) 355-5848
Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>

Google Profile: https://plus.google.com/u/0/109309036874332290399/about

Re: ATS Storage Question

Posted by "Adam W. Dace" <co...@gmail.com>.
No big deal, I was just a bit grumpy that there's no error given.  Happy to
see the documentation is up-to-date.

I think sitting around and simply tuning ATS has made me a bit of a grump
lately.

However, what's really amazing is the speed one can achieve once you get
all the bottlenecks out of the way.
I'm currently tweaking child --> parent failover settings and am kind of in
awe of the speed of this thing.


On Sat, Oct 19, 2013 at 6:16 PM, Alan M. Carroll <
amc@network-geographics.com> wrote:

> Friday, October 18, 2013, 11:30:39 AM, you wrote:
>
> > Even worse, the Admin Guide indicates more than one database file isn't
> currently supported.  Ouch.
>
> That's a bit misleading, it's only one file per physical device. I see
> where that's done but I haven't had time to fix it yet. AFAICT there's no
> real reason for that limitation, it's just an unhappy side effect of how
> the cache spans are calculated.
>
>


-- 
____________________________________________________________
Adam W. Dace <co...@gmail.com>

Phone: (815) 355-5848
Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>

Google Profile: https://plus.google.com/u/0/109309036874332290399/about

Re: ATS Storage Question

Posted by "Alan M. Carroll" <am...@network-geographics.com>.
Friday, October 18, 2013, 11:30:39 AM, you wrote:

> Even worse, the Admin Guide indicates more than one database file isn't currently supported.  Ouch.

That's a bit misleading, it's only one file per physical device. I see where that's done but I haven't had time to fix it yet. AFAICT there's no real reason for that limitation, it's just an unhappy side effect of how the cache spans are calculated.


Re: ATS Storage Question

Posted by "Adam W. Dace" <co...@gmail.com>.
Even worse, the Admin Guide indicates more than one database file isn't
currently supported.  Ouch.

To be honest, I'm already running this and was just questioning my sizing.

Does anyone know the scoop on all this?  I can pretty easily just give up
and live with 2GB of cache.


On Fri, Oct 18, 2013 at 11:20 AM, Reindl Harald <h....@thelounge.net>wrote:

>
>
> Am 18.10.2013 18:14, schrieb Adam W. Dace:
> > Unfortunately, not an option for me
>
> bad
>
> > Although, honestly, I prefer filesystem-based storage anyways
>
> why?
>
> in case of a softeare which can work with RAW devcies you
> have the complete filesystem-overhead including caching and
> so on
>
> ATS raw devices are completly painless and initialized at their own
> even after stop ATS, dd if=/dev/zero of=/dev/sdc bs=1M count=50
> followed by start the service again
>
> service ats stop; dd if=/dev/zero of=/dev/sdc bs=1M count=50; service ats
> start
>
> > I was just wondering if there's a "right answer" here.
> >
> >
> > On Fri, Oct 18, 2013 at 11:09 AM, Reindl Harald <h.reindl@thelounge.net<mailto:
> h.reindl@thelounge.net>> wrote:
> >
> >
> >     Am 18.10.2013 18:01, schrieb Adam W. Dace:
> >     > I've already noticed ATS tends to slow down a bit when using
> database files above 2GB...but it got me thinking.
> >     > Does anyone know which would be more efficient: 2 database files
> at 2GB each...or 4 database files at 1GB each?
> >
> >     no file at all is the solution -> RAW device
> >
> >     [root@proxy:~]$ cat /etc/trafficserver/storage.config
> >     # Storage Configuration file
> >
> >     # /etc/udev/rules.d/50-ats.rules
> >     # Apache Traffic Server owns disk for RAW access
> >     # KERNEL=="sdc", OWNER="ats"
> >
> >     /dev/sdc
>
>


-- 
____________________________________________________________
Adam W. Dace <co...@gmail.com>

Phone: (815) 355-5848
Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>

Google Profile: https://plus.google.com/u/0/109309036874332290399/about

Re: ATS Storage Question

Posted by Reindl Harald <h....@thelounge.net>.

Am 18.10.2013 18:14, schrieb Adam W. Dace:
> Unfortunately, not an option for me

bad

> Although, honestly, I prefer filesystem-based storage anyways

why?

in case of a softeare which can work with RAW devcies you
have the complete filesystem-overhead including caching and
so on

ATS raw devices are completly painless and initialized at their own
even after stop ATS, dd if=/dev/zero of=/dev/sdc bs=1M count=50
followed by start the service again

service ats stop; dd if=/dev/zero of=/dev/sdc bs=1M count=50; service ats start

> I was just wondering if there's a "right answer" here.
> 
> 
> On Fri, Oct 18, 2013 at 11:09 AM, Reindl Harald <h.reindl@thelounge.net <ma...@thelounge.net>> wrote:
> 
> 
>     Am 18.10.2013 18:01, schrieb Adam W. Dace:
>     > I've already noticed ATS tends to slow down a bit when using database files above 2GB...but it got me thinking.
>     > Does anyone know which would be more efficient: 2 database files at 2GB each...or 4 database files at 1GB each?
> 
>     no file at all is the solution -> RAW device
> 
>     [root@proxy:~]$ cat /etc/trafficserver/storage.config
>     # Storage Configuration file
> 
>     # /etc/udev/rules.d/50-ats.rules
>     # Apache Traffic Server owns disk for RAW access
>     # KERNEL=="sdc", OWNER="ats"
> 
>     /dev/sdc


Re: ATS Storage Question

Posted by "Adam W. Dace" <co...@gmail.com>.
Unfortunately, not an option for me.  Although, honestly, I prefer
filesystem-based storage anyways.

I was just wondering if there's a "right answer" here.


On Fri, Oct 18, 2013 at 11:09 AM, Reindl Harald <h....@thelounge.net>wrote:

>
> Am 18.10.2013 18:01, schrieb Adam W. Dace:
> > I've already noticed ATS tends to slow down a bit when using database
> files above 2GB...but it got me thinking.
> > Does anyone know which would be more efficient: 2 database files at 2GB
> each...or 4 database files at 1GB each?
>
> no file at all is the solution -> RAW device
>
> [root@proxy:~]$ cat /etc/trafficserver/storage.config
> # Storage Configuration file
>
> # /etc/udev/rules.d/50-ats.rules
> # Apache Traffic Server owns disk for RAW access
> # KERNEL=="sdc", OWNER="ats"
>
> /dev/sdc
>
>


-- 
____________________________________________________________
Adam W. Dace <co...@gmail.com>

Phone: (815) 355-5848
Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>

Google Profile: https://plus.google.com/u/0/109309036874332290399/about

Re: ATS Storage Question

Posted by Reindl Harald <h....@thelounge.net>.
Am 18.10.2013 18:01, schrieb Adam W. Dace:
> I've already noticed ATS tends to slow down a bit when using database files above 2GB...but it got me thinking.
> Does anyone know which would be more efficient: 2 database files at 2GB each...or 4 database files at 1GB each?

no file at all is the solution -> RAW device

[root@proxy:~]$ cat /etc/trafficserver/storage.config
# Storage Configuration file

# /etc/udev/rules.d/50-ats.rules
# Apache Traffic Server owns disk for RAW access
# KERNEL=="sdc", OWNER="ats"

/dev/sdc


Re: ATS Storage Question

Posted by "Adam W. Dace" <co...@gmail.com>.
Oh ok.  Tomorrow I'll dig up the Firefox extension that lets one measure
load times and see if there's any evidence.

Then I'll file it in Jira.


On Fri, Oct 18, 2013 at 1:16 PM, Adam W. Dace <co...@gmail.com>wrote:

> P.S.  Just FYI, after tuning the rest of the application, the "slowdown"
> I'm experiencing is 5% at best.
> As usual, great job guys.
>
>
> On Fri, Oct 18, 2013 at 1:03 PM, Reindl Harald <h....@thelounge.net>wrote:
>
>>
>>
>> Am 18.10.2013 19:57, schrieb James Peach:
>> > You could also create the files you need, then mount them as raw
>> devices over loopback. That way you get to use files, but ATS treats them
>> as raw devices
>>
>> with what benefit?
>>
>> they additional layer of looback on top of host-OS, guest-OS, guest-FS
>> with all
>> the overhead of each layer? makes pretty no sense outside university
>> studies
>> trying how many layers one can stuck in each other
>>
>>
>
>
> --
> ____________________________________________________________
> Adam W. Dace <co...@gmail.com>
>
> Phone: (815) 355-5848
> Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
> Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>
>
> Google Profile: https://plus.google.com/u/0/109309036874332290399/about
>



-- 
____________________________________________________________
Adam W. Dace <co...@gmail.com>

Phone: (815) 355-5848
Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>

Google Profile: https://plus.google.com/u/0/109309036874332290399/about

Re: ATS Storage Question

Posted by "Adam W. Dace" <co...@gmail.com>.
P.S.  Just FYI, after tuning the rest of the application, the "slowdown"
I'm experiencing is 5% at best.
As usual, great job guys.


On Fri, Oct 18, 2013 at 1:03 PM, Reindl Harald <h....@thelounge.net>wrote:

>
>
> Am 18.10.2013 19:57, schrieb James Peach:
> > You could also create the files you need, then mount them as raw devices
> over loopback. That way you get to use files, but ATS treats them as raw
> devices
>
> with what benefit?
>
> they additional layer of looback on top of host-OS, guest-OS, guest-FS
> with all
> the overhead of each layer? makes pretty no sense outside university
> studies
> trying how many layers one can stuck in each other
>
>


-- 
____________________________________________________________
Adam W. Dace <co...@gmail.com>

Phone: (815) 355-5848
Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>

Google Profile: https://plus.google.com/u/0/109309036874332290399/about

Re: ATS Storage Question

Posted by Leif Hedstrom <zw...@apache.org>.
On Oct 18, 2013, at 11:57 AM, James Peach <jp...@apache.org> wrote:

> On Oct 18, 2013, at 10:19 AM, Leif Hedstrom <zw...@apache.org> wrote:
> 
>> 
>> On Oct 18, 2013, at 11:12 AM, Adam W. Dace <co...@gmail.com> wrote:
>> 
>>> I'm essentially leasing a virtual machine on the cheap, and I didn't see any such option from the company I'm hosting it with.
>> 
>> Does it use LLVM? If so, just carve off a piece of your virtualized disk as a raw device. Raw device here doesn’t imply “physically raw device”, just something that does not have a file system on it.
> 
> You could also create the files you need, then mount them as raw devices over loopback. That way you get to use files, but ATS treats them as raw devices.


That could also be useful as a way to identify where Adam’s slowdown is coming from: if the loopback mounted device does not have the performance issues, then we know it’s entirely ATS’s fault. If it’s the same performance issues as being on the FS directly, it’s due to the kernel / virtualization. 

I’m hoping the loopback test would yield the same results as using a DB file on the filesystem directly :).

— Leif


Re: ATS Storage Question

Posted by Leif Hedstrom <zw...@apache.org>.
On Oct 18, 2013, at 12:10 PM, James Peach <jp...@apache.org> wrote:

> On Oct 18, 2013, at 11:03 AM, Reindl Harald <h....@thelounge.net> wrote:
> 
>> 
>> 
>> Am 18.10.2013 19:57, schrieb James Peach:
>>> You could also create the files you need, then mount them as raw devices over loopback. That way you get to use files, but ATS treats them as raw devices
>> 
>> with what benefit?
>> 
>> they additional layer of looback on top of host-OS, guest-OS, guest-FS with all
>> the overhead of each layer? makes pretty no sense outside university studies
>> trying how many layers one can stuck in each other
> 
> With the benefit that it will work within his virtualized constraints, and it will let him add multiple cache devices. Obviously raw storage would be better, but this will work.


Btw, there is a bug filed on this issue with multiple DBs on the FS. It’s something Alan had noticed as well, and it’s limited to one DB per mount point (I think?).

Fixing this would also be useful, but Adam, please make sure to file appropriate Jira’s with your findings re: performance (I don’t know why a 4GB DB file would be slower than a 2GB file).

— Leif

https://issues.apache.org/jira/browse/TS-1869

Re: ATS Storage Question

Posted by James Peach <jp...@apache.org>.
On Oct 18, 2013, at 11:03 AM, Reindl Harald <h....@thelounge.net> wrote:

> 
> 
> Am 18.10.2013 19:57, schrieb James Peach:
>> You could also create the files you need, then mount them as raw devices over loopback. That way you get to use files, but ATS treats them as raw devices
> 
> with what benefit?
> 
> they additional layer of looback on top of host-OS, guest-OS, guest-FS with all
> the overhead of each layer? makes pretty no sense outside university studies
> trying how many layers one can stuck in each other

With the benefit that it will work within his virtualized constraints, and it will let him add multiple cache devices. Obviously raw storage would be better, but this will work.

J


Re: ATS Storage Question

Posted by Reindl Harald <h....@thelounge.net>.

Am 18.10.2013 19:57, schrieb James Peach:
> You could also create the files you need, then mount them as raw devices over loopback. That way you get to use files, but ATS treats them as raw devices

with what benefit?

they additional layer of looback on top of host-OS, guest-OS, guest-FS with all
the overhead of each layer? makes pretty no sense outside university studies
trying how many layers one can stuck in each other


Re: ATS Storage Question

Posted by James Peach <jp...@apache.org>.
On Oct 18, 2013, at 10:19 AM, Leif Hedstrom <zw...@apache.org> wrote:

> 
> On Oct 18, 2013, at 11:12 AM, Adam W. Dace <co...@gmail.com> wrote:
> 
>> I'm essentially leasing a virtual machine on the cheap, and I didn't see any such option from the company I'm hosting it with.
> 
> Does it use LLVM? If so, just carve off a piece of your virtualized disk as a raw device. Raw device here doesn’t imply “physically raw device”, just something that does not have a file system on it.

You could also create the files you need, then mount them as raw devices over loopback. That way you get to use files, but ATS treats them as raw devices.

J

Re: ATS Storage Question

Posted by "Adam W. Dace" <co...@gmail.com>.
I'll check into that eventually.  Thanks.  For today, I'm so tired of
refreshing my cache I'll accept a little bit of slowdown for 4GB of storage.


On Fri, Oct 18, 2013 at 12:19 PM, Leif Hedstrom <zw...@apache.org> wrote:

>
> On Oct 18, 2013, at 11:12 AM, Adam W. Dace <co...@gmail.com>
> wrote:
>
> > I'm essentially leasing a virtual machine on the cheap, and I didn't see
> any such option from the company I'm hosting it with.
>
> Does it use LLVM? If so, just carve off a piece of your virtualized disk
> as a raw device. Raw device here doesn’t imply “physically raw device”,
> just something that does not have a file system on it.
>
> — Leif
>
>


-- 
____________________________________________________________
Adam W. Dace <co...@gmail.com>

Phone: (815) 355-5848
Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>

Google Profile: https://plus.google.com/u/0/109309036874332290399/about

Re: ATS Storage Question

Posted by Leif Hedstrom <zw...@apache.org>.
On Oct 18, 2013, at 11:12 AM, Adam W. Dace <co...@gmail.com> wrote:

> I'm essentially leasing a virtual machine on the cheap, and I didn't see any such option from the company I'm hosting it with.

Does it use LLVM? If so, just carve off a piece of your virtualized disk as a raw device. Raw device here doesn’t imply “physically raw device”, just something that does not have a file system on it.

— Leif


Re: ATS Storage Question

Posted by Reindl Harald <h....@thelounge.net>.
you only need to get a spare virtual disk to your guest

what you do there is your business and nobody but you has
to decide if you format if with XFS, EXT4 or assign it to ATS

maybe this is really that complicated there
one reason why i always host my own hardware with ESXi/vSphere

Am 18.10.2013 19:12, schrieb Adam W. Dace:
> I'm essentially leasing a virtual machine on the cheap, and I didn't see any such option from the company I'm
> hosting it with.
> 
> On Fri, Oct 18, 2013 at 12:07 PM, Reindl Harald <h.reindl@thelounge.net <ma...@thelounge.net>> wrote:
> 
>     Am 18.10.2013 18:40, schrieb Adam W. Dace:
>     > Yeah, I'm not trying to debate that filesystem is faster than raw disk.  It's just not really an option for me.
>     > Heck, one of the caches I'm running is completely virtualized. Sorry, no raw disk there
> 
>     uhm?
> 
>     [root@proxy:~]$ cat /proc/cpuinfo | grep hypervisor
>     flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2
>     ss syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts nopl xtopology tsc_reliable nonstop_tsc aperfmperf pni
>     pclmulqdq ssse3 cx16 sse4_1 sse4_2 popcnt aes hypervisor lahf_lm ida arat epb pln pts dtherm
> 
>     hypervisor = VMware vSphere 5 guest
> 
>     our whole infrastructure is 100% virtualized including the ATS machine
>     but i see no reason why you can't avoid the guest-FS and assign the
>     virtual disk dedicated to ATS
> 
>     in case of virtualization it is even more important to avoid the
>     additional filesystem layer because there are enough layers with
>     latency (guest, host, SAN)


Re: ATS Storage Question

Posted by "Adam W. Dace" <co...@gmail.com>.
I'm essentially leasing a virtual machine on the cheap, and I didn't see
any such option from the company I'm hosting it with.


On Fri, Oct 18, 2013 at 12:07 PM, Reindl Harald <h....@thelounge.net>wrote:

> Am 18.10.2013 18:40, schrieb Adam W. Dace:
> > Yeah, I'm not trying to debate that filesystem is faster than raw disk.
>  It's just not really an option for me.
> > Heck, one of the caches I'm running is completely virtualized. Sorry, no
> raw disk there
>
> uhm?
>
> [root@proxy:~]$ cat /proc/cpuinfo | grep hypervisor
> flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat
> pse36 clflush dts acpi mmx fxsr sse sse2
> ss syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts nopl xtopology
> tsc_reliable nonstop_tsc aperfmperf pni
> pclmulqdq ssse3 cx16 sse4_1 sse4_2 popcnt aes hypervisor lahf_lm ida arat
> epb pln pts dtherm
>
> hypervisor = VMware vSphere 5 guest
>
> our whole infrastructure is 100% virtualized including the ATS machine
> but i see no reason why you can't avoid the guest-FS and assign the
> virtual disk dedicated to ATS
>
> in case of virtualization it is even more important to avoid the
> additional filesystem layer because there are enough layers with
> latency (guest, host, SAN)
>
>
>
>
>


-- 
____________________________________________________________
Adam W. Dace <co...@gmail.com>

Phone: (815) 355-5848
Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>

Google Profile: https://plus.google.com/u/0/109309036874332290399/about

Re: ATS Storage Question

Posted by "Adam W. Dace" <co...@gmail.com>.
Dang...I retract my question.  A little more experimenting and I
realize...this really -is- unsupported.

Well, I hope someday in the future multiple database files will be
supported.
I think it'd help the performance of my caches.  Thanks all!


On Fri, Oct 18, 2013 at 11:40 AM, Adam W. Dace <co...@gmail.com>wrote:

> Yeah, I'm not trying to debate that filesystem is faster than raw disk.
> It's just not really an option for me.
> Heck, one of the caches I'm running is completely virtualized.  Sorry, no
> raw disk there.
>
> That's why I'm asking about files...did your testing have anything to say
> about one file versus many?
>
>
> On Fri, Oct 18, 2013 at 11:37 AM, Johan Caripson <jo...@director.se>wrote:
>
>> I have done lots of test.
>> Go for raw instead!
>>
>> Now with raw i can push 20 Gbit/s
>> From a single sever with fusion-io drive. With storage files we "only"
>> manage to push around 11 Gbit/s
>>
>> We use 900GB cache storage
>>
>>  Johan Carlsson
>>
>>
>> 2013/10/18 Adam W. Dace <co...@gmail.com>
>>
>> I've already noticed ATS tends to slow down a bit when using database
>>> files above 2GB...but it got me thinking.
>>>
>>> Does anyone know which would be more efficient: 2 database files at 2GB
>>> each...or 4 database files at 1GB each?
>>>
>>> Thanks In Advance,
>>>
>>> --
>>> ____________________________________________________________
>>> Adam W. Dace <co...@gmail.com>
>>>
>>> Phone: (815) 355-5848
>>> Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
>>> Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>
>>>
>>> Google Profile: https://plus.google.com/u/0/109309036874332290399/about
>>>
>>
>>
>
>
> --
> ____________________________________________________________
> Adam W. Dace <co...@gmail.com>
>
> Phone: (815) 355-5848
> Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
> Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>
>
> Google Profile: https://plus.google.com/u/0/109309036874332290399/about
>



-- 
____________________________________________________________
Adam W. Dace <co...@gmail.com>

Phone: (815) 355-5848
Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>

Google Profile: https://plus.google.com/u/0/109309036874332290399/about

Re: ATS Storage Question

Posted by Reindl Harald <h....@thelounge.net>.
Am 18.10.2013 18:40, schrieb Adam W. Dace:
> Yeah, I'm not trying to debate that filesystem is faster than raw disk.  It's just not really an option for me.
> Heck, one of the caches I'm running is completely virtualized. Sorry, no raw disk there

uhm?

[root@proxy:~]$ cat /proc/cpuinfo | grep hypervisor
flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2
ss syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts nopl xtopology tsc_reliable nonstop_tsc aperfmperf pni
pclmulqdq ssse3 cx16 sse4_1 sse4_2 popcnt aes hypervisor lahf_lm ida arat epb pln pts dtherm

hypervisor = VMware vSphere 5 guest

our whole infrastructure is 100% virtualized including the ATS machine
but i see no reason why you can't avoid the guest-FS and assign the
virtual disk dedicated to ATS

in case of virtualization it is even more important to avoid the
additional filesystem layer because there are enough layers with
latency (guest, host, SAN)





Re: ATS Storage Question

Posted by "Adam W. Dace" <co...@gmail.com>.
Yeah, I'm not trying to debate that filesystem is faster than raw disk.
It's just not really an option for me.
Heck, one of the caches I'm running is completely virtualized.  Sorry, no
raw disk there.

That's why I'm asking about files...did your testing have anything to say
about one file versus many?


On Fri, Oct 18, 2013 at 11:37 AM, Johan Caripson <jo...@director.se> wrote:

> I have done lots of test.
> Go for raw instead!
>
> Now with raw i can push 20 Gbit/s
> From a single sever with fusion-io drive. With storage files we "only"
> manage to push around 11 Gbit/s
>
> We use 900GB cache storage
>
> Johan Carlsson
>
>
> 2013/10/18 Adam W. Dace <co...@gmail.com>
>
> I've already noticed ATS tends to slow down a bit when using database
>> files above 2GB...but it got me thinking.
>>
>> Does anyone know which would be more efficient: 2 database files at 2GB
>> each...or 4 database files at 1GB each?
>>
>> Thanks In Advance,
>>
>> --
>> ____________________________________________________________
>> Adam W. Dace <co...@gmail.com>
>>
>> Phone: (815) 355-5848
>> Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
>> Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>
>>
>> Google Profile: https://plus.google.com/u/0/109309036874332290399/about
>>
>
>


-- 
____________________________________________________________
Adam W. Dace <co...@gmail.com>

Phone: (815) 355-5848
Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>

Google Profile: https://plus.google.com/u/0/109309036874332290399/about

Re: ATS Storage Question

Posted by Johan Caripson <jo...@director.se>.
I have done lots of test.
Go for raw instead!

Now with raw i can push 20 Gbit/s
>From a single sever with fusion-io drive. With storage files we "only"
manage to push around 11 Gbit/s

We use 900GB cache storage

Johan Carlsson


2013/10/18 Adam W. Dace <co...@gmail.com>

> I've already noticed ATS tends to slow down a bit when using database
> files above 2GB...but it got me thinking.
>
> Does anyone know which would be more efficient: 2 database files at 2GB
> each...or 4 database files at 1GB each?
>
> Thanks In Advance,
>
> --
> ____________________________________________________________
> Adam W. Dace <co...@gmail.com>
>
> Phone: (815) 355-5848
> Instant Messenger: AIM & Yahoo! IM - colonelforbin74 | ICQ - #39374451
> Microsoft Messenger - colonelforbin74@live.com <ad...@turing.com>
>
> Google Profile: https://plus.google.com/u/0/109309036874332290399/about
>