[9fans] anyone put their venti on an SSD?

[9fans] anyone put their venti on an SSD?

[Steve Simon]

what the subject says, anyone put their venti (those that use it)
on a solid state disk?

-Steve

Re: [9fans] anyone put their venti on an SSD?

[a]

I have not yet, but i've been meaning to play around some with
different arrangements for different parts. Please let us know if
you hear anything interesting.

Anthony

Re: [9fans] anyone put their venti on an SSD?

[geoff]

I tried putting our index on a single OCZ SSD and it died during buildindex.
The SSD was completely unresponsive thereafter, which is pretty
appalling behaviour for a storage device.  Having since sworn off
OCZ, I would try again with a pair of Intel 330s in a RAID.

Re: [9fans] anyone put their venti on an SSD?

[Matthew Veety]

I tried putting venti on an ssd with similar results. Fossil, kenfs, and cwfs all worked fine on that drive though. I think it was one of the earlier Intels.

On May 3, 2013, at 16:59, geoff@plan9.bell-labs.com wrote:

> I tried putting our index on a single OCZ SSD and it died during buildindex.
> The SSD was completely unresponsive thereafter, which is pretty
> appalling behaviour for a storage device.  Having since sworn off
> OCZ, I would try again with a pair of Intel 330s in a RAID.
> 

Re: [9fans] anyone put their venti on an SSD?

[cinap_lenrek]

i have 60GB intel ssd in my new fileserver holding the
cwfs worm cache. no problems so far. but the machine is
just up for two weeks.

its an experiment. if it breaks i have spare sata drive
that could replace it. the worm is on a mirror with
traditional mechanical harddrives.

cpu% cat '#S/sdC0/ctl'
inquiry INTEL SSDSC2CT060A3
config 0040 capabilities 2F00 dma 00550040 dmactl 00550040 rwm 16 rwmctl 0 lba48always off
model	INTEL SSDSC2CT060A3
serial	BTMP31010344060AGN
firm	300i
feat	lba llba smart power nop ata8 sct
geometry 117231408512
alignment 512 0
missirq	0
sloop	0
irq	2233361 2233358
bsy	0 0
nildrive	3
part data 0 117231408
part plan9 63 117226305
part 9fat 63 204863
part nvram 204863 204864
part fscache 204871 67313735

--
cinap

Re: [9fans] anyone put their venti on an SSD?

[Charles Forsyth]

[-- Attachment #1: Type: text/plain, Size: 438 bytes --]

On 3 May 2013 21:59, <geoff@plan9.bell-labs.com> wrote:

> I tried putting our index on a single OCZ SSD and it died during
> buildindex.
> The SSD was completely unresponsive thereafter, which is pretty
> appalling behaviour for a storage device.
>

I had similar problems with OCZ. I was only copying venti state (as raw
partitions) from one drive to the ssd.
I haven't tried again with other makes, mainly for lack of time.

[-- Attachment #2: Type: text/html, Size: 793 bytes --]

Re: [9fans] anyone put their venti on an SSD?

[Lyndon Nerenberg]

On 2013-05-03, at 1:59 PM, geoff@plan9.bell-labs.com wrote:

> I tried putting our index on a single OCZ SSD and it died during buildindex.
> The SSD was completely unresponsive thereafter, which is pretty
> appalling behaviour for a storage device.  Having since sworn off
> OCZ, I would try again with a pair of Intel 330s in a RAID.

A couple of months ago I tried cloning my existing fossil+venti to an OCZ Vertex 3.  I did the partitioning and setup by hand and everything seemed to go just fine: the arenas copied and verified, fossil initialized, and I could run and mount the fossil from the SSD, but whenever I tried to boot from it it failed very early on with "I/O error" and that was that.  I didn't have any spare time to mess around with it so I just stuck with the existing disk.

I was running a really bastardized mix of old and new boot software, so it's quite possible I screwed up installing the correct MBR and boot loader.  But it might also have been a problem with the BIOS or SATA controller on the motherboard -- it's a slightly ancient Supermicro Atom 1U, and it doesn't like SATA hard disks, either (it's currently talking to an IDE disk).  I didn't have time to try 9atom.

That said, the limited testing I did with the secondary fossil+venti showed it to be quite zippy, within the limitations of the surrounding motherboard hardware.  Certainly it was noticeably faster than the hard disk for, e.g., mk all.

--lyndon

Re: [9fans] anyone put their venti on an SSD?

[cinap_lenrek]

ocz seems to have a bad reputation. just googled "intel ssd broken"
and you get tons of results from people with broken/dead ocz ssd's.

--
cinap

Re: [9fans] anyone put their venti on an SSD?

[Devon H. O'Dell]

We've had a lot of success with Intel SSDs, only problem is that they
seem to be in short supply right now. We're also looking at Samsung
SSDs, and they seem to be perhaps even better than the Intel SSDs.
OCZs break often in my experience.

2013/5/3  <cinap_lenrek@gmx.de>:
> ocz seems to have a bad reputation. just googled "intel ssd broken"
> and you get tons of results from people with broken/dead ocz ssd's.
>
> --
> cinap
>

Re: [9fans] anyone put their venti on an SSD?

[erik quanstrom]

> I was running a really bastardized mix of old and new boot software,
> so it's quite possible I screwed up installing the correct MBR and
> boot loader.  But it might also have been a problem with the BIOS or
> SATA controller on the motherboard -- it's a slightly ancient
> Supermicro Atom 1U, and it doesn't like SATA hard disks, either (it's
> currently talking to an IDE disk).  I didn't have time to try 9atom.

that sort of machine is 9atom's original raison d'être.
also there were some issues with ide in pio mode recently
fixed.  (credit: stallion)  you may still want to look at 9atom
even in ide mode.

(i'm not absolving the drive.  just saying that there are plan 9
issues affecting your machine.)

- erik

Re: [9fans] anyone put their venti on an SSD?

[Lyndon Nerenberg]

On 2013-05-03, at 6:31 PM, cinap_lenrek@gmx.de wrote:

> ocz seems to have a bad reputation. just googled "intel ssd broken"
> and you get tons of results from people with broken/dead ocz ssd's.

Disk drive reliability comes and goes with the seasons.  For years I only ran Seagate disks, and wouldn't go near WD.  Then, after a 30% failure run on < 1 year old Seagates, I switched back to WDs, which have been flawless for me.  So far.  And Hitachi has drifted in and out of the picture over the years.

Ask anyone else and they will tell you a partially to completely different story.  At $80 for a couple of TB, I'll just toss them out when they break.  And while SSD isn't anywhere near that inexpensive yet, it's finally down into the price range where you can call it commodity disk.  These days I'm paying a bit under $1/GB for SSD.  That's cheap enough that I can afford to experiment with smaller sized drives in non-critical applications.

I have a FreeBSD box that's been running off an OCZ Vertex 3 for a few months now without issue. It does regular buildworlds and the like, so I'm not gentle on the write volume.  UFS TRIM support seems to work fine, and if you chose to believe the smart stats, the (128 GB) disk is error free after having many TB written to it.  Many claim Sandforce disks will blow up well before this.  I'm beginning to think otherwise, although my sample space is not statistically significant.

I have a pair of Vertex 4s I haven't installed yet.  I'm curious to see how much (or if) the new OCZ controllers work.  The benchmarks I've read are impressive.  I'm planning to put one into my Mac Mini.  If MacOS and its several hundred GB VM footprint can't kill an SSD, nothing will ;-)

--lyndon

Re: [9fans] anyone put their venti on an SSD?

[Lyndon Nerenberg]

On 2013-05-03, at 6:43 PM, erik quanstrom wrote:

> (i'm not absolving the drive.  just saying that there are plan 9
> issues affecting your machine.)

No doubt.  But I got it to the point where it's working quite happily.  And it can maintain that steady state until it tips over and dies, at which point I will re-install on much more current kit.  I'm getting too old to fight with the hardware any more ;-P

Re: [9fans] anyone put their venti on an SSD?

[erik quanstrom]

> Disk drive reliability comes and goes with the seasons.  For years I
> only ran Seagate disks, and wouldn't go near WD. Then, after a 30%
> failure run on < 1 year old Seagates, I switched back to WDs, which
> have been flawless for me.  So far.  And Hitachi has drifted in and
> out of the picture over the years.

imho, applying 50 years of experience with spinning hard drives with
the relatively new flash memory drive is a suspect comparison.

- erik

Re: [9fans] anyone put their venti on an SSD?

[Lyndon Nerenberg]

On 2013-05-03, at 6:51 PM, erik quanstrom wrote:

> imho, applying 50 years of experience with spinning hard drives with
> the relatively new flash memory drive is a suspect comparison.

Is it?  Cheap SSD seems to break as often as cheap spinny disks.  According to everyone's anecdotal stories, at least.

Re: [9fans] anyone put their venti on an SSD?

[erik quanstrom]

On Fri May  3 17:59:27 EDT 2013, cinap_lenrek@gmx.de wrote:
> i have 60GB intel ssd in my new fileserver holding the
> cwfs worm cache. no problems so far. but the machine is
> just up for two weeks.

as long as we're straying from venti, i'll say that i've used ssds
in ken's file server as both cache and worm drives.  there
were no problems at all.  i've mostly used various intel drives,
which have performed in accordance to all published specs.

i've also used ocz drives as boot drives to hold nvram.  no
problems.  perhaps this is the proverbial faint praise.

on the subject of ken's file server / cwfs, it makes the most sense
to me to use ssds as worm, not cache.  bandwith to the cache
isn't important as long as you don't blow the ram cache.  but
anything that isn't in the ram cache but is on the worm tends
to be randomish access.  ssds to very well at at random access,
and write once decreases pressure on many things that make
implementing a flash drive hard.

i would expect similar from venti, which has an even more random
data access pattern.

- erik

Re: [9fans] anyone put their venti on an SSD?

[erik quanstrom]

> On 2013-05-03, at 6:51 PM, erik quanstrom wrote:
>
> > imho, applying 50 years of experience with spinning hard drives with
> > the relatively new flash memory drive is a suspect comparison.
>
> Is it?  Cheap SSD seems to break as often as cheap spinny disks.  According to everyone's anecdotal stories, at least

well clearly, we must lump everything
that breaks "anecdotally as often"
in the same catagory,
by manufacturer.

- erik

Re: [9fans] anyone put their venti on an SSD?

[Lyndon Nerenberg]

On 2013-05-03, at 7:00 PM, erik quanstrom wrote:

> well clearly, we must lump everything
> that breaks "anecdotally as often"
> in the same catagory,
> by manufacturer.

Exactly.  That's where we started this conversation :-)

Re: [9fans] anyone put their venti on an SSD?

[Kurt H Maier]

On Fri, May 03, 2013 at 10:00:41PM -0400, erik quanstrom wrote:
> > On 2013-05-03, at 6:51 PM, erik quanstrom wrote:
> >
> > > imho, applying 50 years of experience with spinning hard drives with
> > > the relatively new flash memory drive is a suspect comparison.
> >
> > Is it?  Cheap SSD seems to break as often as cheap spinny disks.  According to everyone's anecdotal stories, at least
>
> well clearly, we must lump everything
> that breaks "anecdotally as often"
> in the same catagory,
> by manufacturer.
>
> - erik
>

that was the worst haiku I've ever seen

Re: [9fans] anyone put their venti on an SSD?

[erik quanstrom]

> > well clearly, we must lump everything
> > that breaks "anecdotally as often"
> > in the same catagory,
> > by manufacturer.
> >
> > - erik
> >
>
> that was the worst haiku I've ever seen

oh, now.
you give me too much credit.
i wasn't even
trying.

- erik

Re: [9fans] anyone put their venti on an SSD?

[Steven Stallion]

[-- Attachment #1: Type: text/plain, Size: 540 bytes --]

I had originally used a Crucial 32GB SSD years ago and swapped to a 55GB
OCZ enterprise drive (using sdahci). More recently I've moved my venti
arenas over to plan9port and have switched over to using the entire SSD for
fossil. So far this has been faster than running venti natively - though I
still take a replica each night out of paranoia.

On Fri, May 3, 2013 at 7:19 AM, Steve Simon <steve@quintile.net> wrote:

> what the subject says, anyone put their venti (those that use it)
> on a solid state disk?
>
> -Steve
>
>

[-- Attachment #2: Type: text/html, Size: 903 bytes --]

Re: [9fans] anyone put their venti on an SSD?

[Steven Stallion]

[-- Attachment #1.1: Type: text/plain, Size: 1517 bytes --]

Agreed. There were a couple of surprises in sdata/sdide with some of the
PIO timings.

A bit of a tangeng, but a while back I switched my Plan 9 machines over to
Atom and haven't looked back. The current configuration I use has a pair of
D525 boards with 4GB of DDR3 and a small SATADOM for nvram/9fat. The file
server has a couple of 2.5" trays I use for write cache and a StarTech.com
PCIe 8-port serial card (patch/uart-ox). I've built close to 20 of these
over the last couple of years, mostly as standalone console servers. I've
had good luck with the OCZ's, though I avoid consumer grade drives like the
plague.

Obligatory picture attached.


On Fri, May 3, 2013 at 6:43 PM, erik quanstrom <quanstro@quanstro.net>wrote:

> > I was running a really bastardized mix of old and new boot software,
> > so it's quite possible I screwed up installing the correct MBR and
> > boot loader.  But it might also have been a problem with the BIOS or
> > SATA controller on the motherboard -- it's a slightly ancient
> > Supermicro Atom 1U, and it doesn't like SATA hard disks, either (it's
> > currently talking to an IDE disk).  I didn't have time to try 9atom.
>
> that sort of machine is 9atom's original raison d'être.
> also there were some issues with ide in pio mode recently
> fixed.  (credit: stallion)  you may still want to look at 9atom
> even in ide mode.
>
> (i'm not absolving the drive.  just saying that there are plan 9
> issues affecting your machine.)
>
> - erik
>
>

[-- Attachment #1.2: Type: text/html, Size: 2043 bytes --]

[-- Attachment #2: IMG_0116.jpg --]
[-- Type: image/jpeg, Size: 1864978 bytes --]

Re: [9fans] anyone put their venti on an SSD?

[cinap_lenrek]

cwfs copies the blocks from worm into the cache on read.
so the working set is served from the ssd and the ram
buffer cache. reading /n/dump would hit the mechanical
disk tho.

--
cinap

Re: [9fans] anyone put their venti on an SSD?

[cinap_lenrek]

no cats in picture!

--
cinap

Re: [9fans] anyone put their venti on an SSD?

[Lyndon Nerenberg]

[-- Attachment #1: Type: text/plain, Size: 94 bytes --]

On 2013-05-03, at 9:19 PM, cinap_lenrek@gmx.de wrote:

> no cats in picture!

Wrong.



[-- Attachment #2: Cat_BandidoI.jpg --]
[-- Type: image/jpg, Size: 51400 bytes --]

[-- Attachment #3: Type: text/plain, Size: 4 bytes --]

Re: [9fans] anyone put their venti on an SSD?

[Charles Forsyth]

[-- Attachment #1: Type: text/plain, Size: 270 bytes --]

On 4 May 2013 00:59, Charles Forsyth <charles.forsyth@gmail.com> wrote:

> I had similar problems with OCZ.


I ought to have observed that I bought the drive from Fry's as refurbished,
which probably wasn't a good recommendation for a drive that was fairly new.

[-- Attachment #2: Type: text/html, Size: 571 bytes --]

Re: [9fans] anyone put their venti on an SSD?

[erik quanstrom]

On Sat May  4 00:18:46 EDT 2013, cinap_lenrek@gmx.de wrote:
> cwfs copies the blocks from worm into the cache on read.
> so the working set is served from the ssd and the ram
> buffer cache. reading /n/dump would hit the mechanical
> disk tho.

that's an option for ken's fs.  i haven't found that it's fast
enough because the latency to retrieve a block that's not
in the cache is the latency for read(worm)+read(cachebkt)+
write(cache).  if one keeps the cache small, the read of
the cachebkt goes away.  therefore it's best to keep
cachesize << wormsize.

- erik

Re: [9fans] anyone put their venti on an SSD?

[Bakul Shah]

> > I had similar problems with OCZ.
>
> I ought to have observed that I bought the drive from Fry's as refurbished,
> which probably wasn't a good recommendation for a drive that was fairly new.

Three strikes. Fate couldn't resist.

Re: [9fans] anyone put their venti on an SSD?

[erik quanstrom]

On Fri May  3 23:27:40 EDT 2013, sstallion@gmail.com wrote:

> I had originally used a Crucial 32GB SSD years ago and swapped to a 55GB
> OCZ enterprise drive (using sdahci). More recently I've moved my venti
> arenas over to plan9port and have switched over to using the entire SSD for
> fossil. So far this has been faster than running venti natively - though I
> still take a replica each night out of paranoia.

there are many components to the performance here, but i can think of at
least one simple issue.  plan 9 loopback tcp can be pretty slow due to scheduling
issues.  (induced by the structure of tcp, not any issue with the scheduler.)
my atom (which is slightly worse than a d525) gets

	ladd; nettest -l -n 10000
	tcp!192.168.0.136!39769 count 10000; 81920000 bytes in 2.649575 s @ 29.5 MB/s (0ms)

so i would imagine that working out how to post a fd to srv would really help
a lot.  not to mention, taking some funk out of configuring the network on
boot.  :-)

- erik

Re: [9fans] anyone put their venti on an SSD?

[Steven Stallion]

[-- Attachment #1: Type: text/plain, Size: 1324 bytes --]

Makes sense. Moving to plan9port had more to do with making better use of
the ReadyNAS in the rack than anything else. The performance was a nice if
unexpected side-effect.

On Friday, May 3, 2013, erik quanstrom wrote:

> On Fri May  3 23:27:40 EDT 2013, sstallion@gmail.com <javascript:;> wrote:
>
> > I had originally used a Crucial 32GB SSD years ago and swapped to a 55GB
> > OCZ enterprise drive (using sdahci). More recently I've moved my venti
> > arenas over to plan9port and have switched over to using the entire SSD
> for
> > fossil. So far this has been faster than running venti natively - though
> I
> > still take a replica each night out of paranoia.
>
> there are many components to the performance here, but i can think of at
> least one simple issue.  plan 9 loopback tcp can be pretty slow due to
> scheduling
> issues.  (induced by the structure of tcp, not any issue with the
> scheduler.)
> my atom (which is slightly worse than a d525) gets
>
>         ladd; nettest -l -n 10000
>         tcp!192.168.0.136!39769 count 10000; 81920000 bytes in 2.649575 s
> @ 29.5 MB/s (0ms)
>
> so i would imagine that working out how to post a fd to srv would really
> help
> a lot.  not to mention, taking some funk out of configuring the network on
> boot.  :-)
>
> - erik
>
>

[-- Attachment #2: Type: text/html, Size: 1589 bytes --]