Mailing List Archive

>i was wonderin, what kind of partition + fs table would be
optimal on server and/or desktop.
What are you planning on using the computer for? And when will you
need the rapid disk access that raid0 gives you?
--DA


--
gentoo-performance@gentoo.org mailing list

Daniel Armyr kirjoitti:
>> i was wonderin, what kind of partition + fs table would be
>>
> optimal on server and/or desktop.
> What are you planning on using the computer for? And when will you
> need the rapid disk access that raid0 gives you?
> --DA
>
>
>
Well, it is just for sake of wondering, and it would not hurt to have
the speediest system ever, even if i would ever need it. That
server/desktop part means that i have desktop computers and one home
server. It also means that id like to know if the setup would be
different. The kind of stuff im wondering is, what should reside on
separate disks to receive any system speedup, even in theory.
--
gentoo-performance@gentoo.org mailing list

I recommend using your home server as a build server and hosting the
portage tree there for the other clients to (auto)mount via NFS.
There's some docs in the wiki on how to do this. It will free you of
the need to be concerned about a separate portage partition on each of
these other machines.

Brian

Mikko Husari wrote:
> Daniel Armyr kirjoitti:
>>> i was wonderin, what kind of partition + fs table would be
>>>
>> optimal on server and/or desktop. What are you planning on using the
>> computer for? And when will you
>> need the rapid disk access that raid0 gives you? --DA
>>
>>
>>
> Well, it is just for sake of wondering, and it would not hurt to have
> the speediest system ever, even if i would ever need it. That
> server/desktop part means that i have desktop computers and one home
> server. It also means that id like to know if the setup would be
> different. The kind of stuff im wondering is, what should reside on
> separate disks to receive any system speedup, even in theory.
--
gentoo-performance@gentoo.org mailing list

Brian Kroth wrote:
> I recommend using your home server as a build server and hosting the
> portage tree there for the other clients to (auto)mount via NFS.
> There's some docs in the wiki on how to do this. It will free you of
> the need to be concerned about a separate portage partition on each of
> these other machines.
>
> Brian
>
> Mikko Husari wrote:
>> Daniel Armyr kirjoitti:
>>>> i was wonderin, what kind of partition + fs table would be
>>>>
>>> optimal on server and/or desktop. What are you planning on using the
>>> computer for? And when will you
>>> need the rapid disk access that raid0 gives you? --DA
>>>
>>>
>>>
>> Well, it is just for sake of wondering, and it would not hurt to have
>> the speediest system ever, even if i would ever need it. That
>> server/desktop part means that i have desktop computers and one home
>> server. It also means that id like to know if the setup would be
>> different. The kind of stuff im wondering is, what should reside on
>> separate disks to receive any system speedup, even in theory.
well, my experiences with distcc has not been quite nice, some of the
packages does not behave well with distributed compiling... or what did
you mean with buildsystem? and searching from portage over nfs is much
slower than searching local portage tree...
--
gentoo-performance@gentoo.org mailing list

>and searching from portage over nfs is much slower than searching
local portage tree...

True, but then you allways do all searches using esearch, (or eix,
but I never used that) and in that case you do the search from
some form of preindexed database that resides locally.

As to the original question, a raid0 will naturally be
theoretically faster when reading several small files
simultaneously. But depending on bandwidth of your HD-controller,
writing might (in theory) be slower if you don't have
HW-accellerated RAID.
--DA


--
gentoo-performance@gentoo.org mailing list

> well, my experiences with distcc has not been quite nice, some of the
>
> packages does not behave well with distributed compiling... or what
> did
> you mean with buildsystem? and searching from portage over nfs is much
>
> slower than searching local portage tree...

As someone else said esearch, eix are very nice tools that work best for searching even when you have a local portage tree.

What I actually meant by build system is to only build packages on one machine and let your other hosts use those binary packages rather than trying to build and compile everywhere. If you want you can use distcc on top of that but it's not necessary. This was more so in relation to your home networking proposal. It results in a single build environment so all of your machines that use those packages behave the same and there's only one machine doing portage tree syncing which means less disk trashing and (external) network traffic. In my experience it ends up performing much better and being easier to manage. Then you can dedicate your other machines to their actual purposes rather than trying to fix build problems in many different places.

As to your original question, again, it depends on what you're doing. RAID0 is best for fast writes in which you don't care about reliability - basically scratch space. Something like portage might be suited for this. Then again, the portage tree is really meant as a local cache of the master tree. As such its supposed to be read from more often than written to. RAID1 provides reliability and read performance since you can read from one of several disks to achieve the same results. Write performance for RAID1 is obviously not as good, but how often are you really writing to /opt or /usr for instance. If you're going software raid, both of these come almost free in terms of overhead involved. With higher levels you'll definitely want true hardware level raid, not some cheap BIOS implemented version. I'd previously read, and can't find the document just now, that RAID10 offers the combination of both of these and the best performance (better than RAID5 as well). Thou
gh that may be out of the question in terms of the number of disks involved.

Here's what I've done in the past for client machines...

Partition1 RAID1 15G /
Partition2 1xRAM/disk swap (it should automatically be striped)
Partition3 RAID0 5G /tmp (you may want more if you're doing video editing or something like that)
Partition4 (extended)
Partition5 RAID1 5G /var (you may want more if you're building packages or something)
Partition6 RAID1 * /home (the rest of it)

Also, if you want to be able to tweak your partition sizes, LVM offers striping and mirroring so you don't need to layer software raid on top of or underneath that as well.

Hope that helps,
Brian
--
gentoo-performance@gentoo.org mailing list

BRIAN PAUL KROTH wrote:
>> well, my experiences with distcc has not been quite nice, some of the
>>
>> packages does not behave well with distributed compiling... or what
>> did
>> you mean with buildsystem? and searching from portage over nfs is much
>>
>> slower than searching local portage tree...
>>
>
> As someone else said esearch, eix are very nice tools that work best for searching even when you have a local portage tree.
>
> What I actually meant by build system is to only build packages on one machine and let your other hosts use those binary packages rather than trying to build and compile everywhere. If you want you can use distcc on top of that but it's not necessary. This was more so in relation to your home networking proposal. It results in a single build environment so all of your machines that use those packages behave the same and there's only one machine doing portage tree syncing which means less disk trashing and (external) network traffic. In my experience it ends up performing much better and being easier to manage. Then you can dedicate your other machines to their actual purposes rather than trying to fix build problems in many different places.
>
> As to your original question, again, it depends on what you're doing. RAID0 is best for fast writes in which you don't care about reliability - basically scratch space. Something like portage might be suited for this. Then again, the portage tree is really meant as a local cache of the master tree. As such its supposed to be read from more often than written to. RAID1 provides reliability and read performance since you can read from one of several disks to achieve the same results. Write performance for RAID1 is obviously not as good, but how often are you really writing to /opt or /usr for instance. If you're going software raid, both of these come almost free in terms of overhead involved. With higher levels you'll definitely want true hardware level raid, not some cheap BIOS implemented version. I'd previously read, and can't find the document just now, that RAID10 offers the combination of both of these and the best performance (better than RAID5 as well). Thou
> gh that may be out of the question in terms of the number of disks involved.
>
> Here's what I've done in the past for client machines...
>
> Partition1 RAID1 15G /
> Partition2 1xRAM/disk swap (it should automatically be striped)
> Partition3 RAID0 5G /tmp (you may want more if you're doing video editing or something like that)
> Partition4 (extended)
> Partition5 RAID1 5G /var (you may want more if you're building packages or something)
> Partition6 RAID1 * /home (the rest of it)
>
> Also, if you want to be able to tweak your partition sizes, LVM offers striping and mirroring so you don't need to layer software raid on top of or underneath that as well.
>
> Hope that helps,
> Brian
>
well, all of this has been really helpfull. my current plans for my
desktop-pc:

MOUNT PSIZE MSIZE FSTYPE BLOCKS RAIDLEVEL
/ 5G 5G xfs 4096 1
/home * * xfs 4096 1
/var 5G 5G xfs 4096 1
/usr 6G 6G xfs 4096 1
/usr/portage 500 1G reiser 2048 0
/usr/portage/distfiles 500 1G xfs 4096 0
/tmp 2G 4G reiser 4096 0


im just not so sure about the filesystem choices...

--
gentoo-performance@gentoo.org mailing list

I've found XFS to be a little disappointing for day to day sort of
stuff. I use ReiserFS almost exclusively, though I have seen how XFS
out performs it in large file related things. In general, most system
files are small (a couple of MB max) as are a lot of user files. You
can do a comparison on your own system with a simple find command.
Perhaps a media volume with XFS would make sense, but then again you're
probably reading it mostly and not writing too much.

I use ext2 for any filesystem where you can avoid the overhead of
journaling if it's not required, like /tmp. One downside I've noticed
with ReiserFS is the long mount time for large (~500G) volumes.

I don't have any numbers for any of this right now, and am often
skeptical of those you see in various articles or even Hans's own site.
I've been planning on doing my own survey including Reiser4 once it
comes out and how they may relate to RAID configurations as well. So
far, this is really just based on some simple bonnie tests and
experience with my servers and workstations at work and home over the
past couple of years.

Also, if the tree grows much more, or whether or not you want a
distfiles cache, you may want to increase the size of those portage
partitions.

Brian

Mikko Husari wrote:
> BRIAN PAUL KROTH wrote:
>>> well, my experiences with distcc has not been quite nice, some of the
>>>
>>> packages does not behave well with distributed compiling... or what
>>> did
>>> you mean with buildsystem? and searching from portage over nfs is much
>>>
>>> slower than searching local portage tree...
>>>
>> As someone else said esearch, eix are very nice tools that work best for searching even when you have a local portage tree.
>>
>> What I actually meant by build system is to only build packages on one machine and let your other hosts use those binary packages rather than trying to build and compile everywhere. If you want you can use distcc on top of that but it's not necessary. This was more so in relation to your home networking proposal. It results in a single build environment so all of your machines that use those packages behave the same and there's only one machine doing portage tree syncing which means less disk trashing and (external) network traffic. In my experience it ends up performing much better and being easier to manage. Then you can dedicate your other machines to their actual purposes rather than trying to fix build problems in many different places.
>>
>> As to your original question, again, it depends on what you're doing. RAID0 is best for fast writes in which you don't care about reliability - basically scratch space. Something like portage might be suited for this. Then again, the portage tree is really meant as a local cache of the master tree. As such its supposed to be read from more often than written to. RAID1 provides reliability and read performance since you can read from one of several disks to achieve the same results. Write performance for RAID1 is obviously not as good, but how often are you really writing to /opt or /usr for instance. If you're going software raid, both of these come almost free in terms of overhead involved. With higher levels you'll definitely want true hardware level raid, not some cheap BIOS implemented version. I'd previously read, and can't find the document just now, that RAID10 offers the combination of both of these and the best performance (better than RAID5 as well). Th
ou
>> gh that may be out of the question in terms of the number of disks involved.
>>
>> Here's what I've done in the past for client machines...
>>
>> Partition1 RAID1 15G /
>> Partition2 1xRAM/disk swap (it should automatically be striped)
>> Partition3 RAID0 5G /tmp (you may want more if you're doing video editing or something like that)
>> Partition4 (extended)
>> Partition5 RAID1 5G /var (you may want more if you're building packages or something)
>> Partition6 RAID1 * /home (the rest of it)
>>
>> Also, if you want to be able to tweak your partition sizes, LVM offers striping and mirroring so you don't need to layer software raid on top of or underneath that as well.
>>
>> Hope that helps,
>> Brian
>>
> well, all of this has been really helpfull. my current plans for my
> desktop-pc:
>
> MOUNT PSIZE MSIZE FSTYPE BLOCKS RAIDLEVEL
> / 5G 5G xfs 4096 1
> /home * * xfs 4096 1
> /var 5G 5G xfs 4096 1
> /usr 6G 6G xfs 4096 1
> /usr/portage 500 1G reiser 2048 0
> /usr/portage/distfiles 500 1G xfs 4096 0
> /tmp 2G 4G reiser 4096 0
>
>
> im just not so sure about the filesystem choices...
>
> --
> gentoo-performance@gentoo.org mailing list
>
--
gentoo-performance@gentoo.org mailing list

Mikko Husari wrote:
> well, all of this has been really helpfull. my current plans for my
> desktop-pc:
>
> MOUNT PSIZE MSIZE FSTYPE BLOCKS RAIDLEVEL
> / 5G 5G xfs 4096 1
> /home * * xfs 4096 1
> /var 5G 5G xfs 4096 1
> /usr 6G 6G xfs 4096 1
> /usr/portage 500 1G reiser 2048 0
> /usr/portage/distfiles 500 1G xfs 4096 0
> /tmp 2G 4G reiser 4096 0
>
>
> im just not so sure about the filesystem choices...
>
>
I would use reiserfs on /var either, for there might be a lot small
files (e.g. squid cache).



___________________________________________________________
To help you stay safe and secure online, we've developed the all new Yahoo! Security Centre. http://uk.security.yahoo.com
--
gentoo-performance@gentoo.org mailing list