|
Home > Archive > Data Storage > June 2006 > CF cards vs. SCSI RAID
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
CF cards vs. SCSI RAID
|
|
|
| We have some servers like DNS that are very important, and have very
low disk usage and access. We currently use 2 17Gb SCSI 1 for them, but
it is hard to find replacement disks and every new disk our people find
is at least 36Gb.
I think it is a waste to have 36Gb on a server that will never use more
than 400mb and I was reading that CFs are more durable and reliable
than hard disks.
We are looking to buy some new supermicro servers and I noticed that
some 1U servers have IDE CF Card support.
So, my question is: is it a good idea to replace those SCSI RAID 1 with
simple 1or 2 Gb CF cards? Is it reliable against RAID 1?
And what about logs? Would I need to control how linux writes the log
or today's CF cards are prepared to this kind of work?
Thanks in advance.
| |
| robertwessel2@yahoo.com 2006-06-26, 1:12 pm |
|
Natan wrote:
> We have some servers like DNS that are very important, and have very
> low disk usage and access. We currently use 2 17Gb SCSI 1 for them, but
> it is hard to find replacement disks and every new disk our people find
> is at least 36Gb.
>
> I think it is a waste to have 36Gb on a server that will never use more
> than 400mb and I was reading that CFs are more durable and reliable
> than hard disks.
>
> We are looking to buy some new supermicro servers and I noticed that
> some 1U servers have IDE CF Card support.
>
> So, my question is: is it a good idea to replace those SCSI RAID 1 with
> simple 1or 2 Gb CF cards? Is it reliable against RAID 1?
>
> And what about logs? Would I need to control how linux writes the log
> or today's CF cards are prepared to this kind of work?
While I don't have a hard study to cite, from experience, and what I've
heard, the consumer grade flash devices appear to be horribly
unreliable as compared to a server grade disk drive. There are
industrial/high reliability CF devices which claim quite high MTBFs
(eg. 2 million hours), but these are rather more expensive, and you'll
probably still need a pair in a a RAID 1 configuration to get into the
same reliability range as you've got now.
There would be no particular reason this wouldn't work. Most OS's can
boot off a flash drive, although it can be tough getting anything other
than FAT on the device (less an issue with Linux). Flash devices do
have the write cycle problem, so if you do many writes you many want to
reconsider, or provide enough extra capacity so that wear leveling can
compensate (I would not expect that to be an issue for most DNS
servers). And then there's the performance issue. Flash drives have
decent random access characteristics (eg. "seek" times), but terrible
data rates, especially on writes (again, probably not an issue for a
DNS server, except for boot times).
That being said, there are certainly plenty of appliances (including
DNS appliances) that boot from flash.
All that to replace a pair of $200 disk drives?
| |
| Faeandar 2006-06-26, 7:12 pm |
| On 26 Jun 2006 08:36:53 -0700, "robertwessel2@yahoo.com"
<robertwessel2@yahoo.com> wrote:
>
>Natan wrote:
>
>
>While I don't have a hard study to cite, from experience, and what I've
>heard, the consumer grade flash devices appear to be horribly
>unreliable as compared to a server grade disk drive. There are
>industrial/high reliability CF devices which claim quite high MTBFs
>(eg. 2 million hours), but these are rather more expensive, and you'll
>probably still need a pair in a a RAID 1 configuration to get into the
>same reliability range as you've got now.
>
>There would be no particular reason this wouldn't work. Most OS's can
>boot off a flash drive, although it can be tough getting anything other
>than FAT on the device (less an issue with Linux). Flash devices do
>have the write cycle problem, so if you do many writes you many want to
>reconsider, or provide enough extra capacity so that wear leveling can
>compensate (I would not expect that to be an issue for most DNS
>servers). And then there's the performance issue. Flash drives have
>decent random access characteristics (eg. "seek" times), but terrible
>data rates, especially on writes (again, probably not an issue for a
>DNS server, except for boot times).
>
>That being said, there are certainly plenty of appliances (including
>DNS appliances) that boot from flash.
>
>All that to replace a pair of $200 disk drives?
Exactly. If you really want to replace internal drives why not do
iSCSI boot/OS from a larger RAID system already in use?
~F
| |
| Morten Reistad 2006-06-26, 7:12 pm |
| In article <1151336212.905809.213100@r2g2000cwb.googlegroups.com>,
robertwessel2@yahoo.com <robertwessel2@yahoo.com> wrote:
>
>Natan wrote:
>
>
>While I don't have a hard study to cite, from experience, and what I've
>heard, the consumer grade flash devices appear to be horribly
>unreliable as compared to a server grade disk drive. There are
>industrial/high reliability CF devices which claim quite high MTBFs
>(eg. 2 million hours), but these are rather more expensive, and you'll
>probably still need a pair in a a RAID 1 configuration to get into the
>same reliability range as you've got now.
We use these for reliability; not cost. Also, bringing down the
power budget is an issue.
We use ATA adapters, so they appear like normal ATA devices. They
can be exposed to volume management and mirroring just like any other
device.
We are extremely concervative with the write cycles to them. We
mount root ro, and have a ramdisk var. The advantage is that
you can, in fact, powercycle the box without any ill effects.
>There would be no particular reason this wouldn't work. Most OS's can
>boot off a flash drive, although it can be tough getting anything other
>than FAT on the device (less an issue with Linux). Flash devices do
The devices format into any file system you like, just as a disk does.
I even have one formatted as a DB9 file system.
>have the write cycle problem, so if you do many writes you many want to
>reconsider, or provide enough extra capacity so that wear leveling can
>compensate (I would not expect that to be an issue for most DNS
>servers). And then there's the performance issue. Flash drives have
>decent random access characteristics (eg. "seek" times), but terrible
>data rates, especially on writes (again, probably not an issue for a
>DNS server, except for boot times).
>
>That being said, there are certainly plenty of appliances (including
>DNS appliances) that boot from flash.
>
>All that to replace a pair of $200 disk drives?
All that to have no moving parts.
-- mrr
| |
| robertwessel2@yahoo.com 2006-06-27, 1:12 am |
|
Morten Reistad wrote:
> We use these for reliability; not cost. Also, bringing down the
> power budget is an issue.
>
> We use ATA adapters, so they appear like normal ATA devices. They
> can be exposed to volume management and mirroring just like any other
> device.
>
> We are extremely concervative with the write cycles to them. We
> mount root ro, and have a ramdisk var. The advantage is that
> you can, in fact, powercycle the box without any ill effects.
> (...)
> All that to have no moving parts.
The lack of moving parts is a much bigger issue when a box is not in a
computer room like the typical DNS server. As I mentioned, I doubt the
consumer grade flash devices are anywhere near as reliable as the
typical server class SCSI drive, and even then expensive
industrial/critical use grade flash devices are only in the same class.
So for a stationary system, I don't think there are any reliability
benefits.
You're absolutely right on power and heat, though.
| |
|
| Yes, it's not about the cost. There is power saving, heat, space... but
my main concern was about reliability.
I thought that could flash discs last much longer without failing than
hard drives, so it would mean less maintenance (not that SCSI is bad,
they last ~3 years here).
And although flash discs have limited write cycles, mounting /var and
/tmp in ram disk is a great option. Even modifying /etc every some
times a week, I believe this could last a decade before starting having
problems.
But it was just an idea. I'll probably test this a lot before thinking
about production.
Thank you for your comments!
robertwessel2@yahoo.com escreveu:
> Morten Reistad wrote:
>
>
> The lack of moving parts is a much bigger issue when a box is not in a
> computer room like the typical DNS server. As I mentioned, I doubt the
> consumer grade flash devices are anywhere near as reliable as the
> typical server class SCSI drive, and even then expensive
> industrial/critical use grade flash devices are only in the same class.
> So for a stationary system, I don't think there are any reliability
> benefits.
>
> You're absolutely right on power and heat, though.
| |
| Thor Lancelot Simon 2006-06-28, 1:14 pm |
| In article <1151370106.321739.254800@i40g2000cwc.googlegroups.com>,
robertwessel2@yahoo.com <robertwessel2@yahoo.com> wrote:
>
>The lack of moving parts is a much bigger issue when a box is not in a
>computer room like the typical DNS server. As I mentioned, I doubt the
>consumer grade flash devices are anywhere near as reliable as the
>typical server class SCSI drive
You're a fool, or you've been drinking too much of some fool's kool-aid.
Semiconductor devices are orders of magnitude more reliable than anything
with moving parts in it -- and always will be.
The issue with flash is, of course, write cycles. So people whose systems
boot and run from flash for reliability reasons commonly take simple
measures to ensure that the flash is not written in normal system operation.
Do you actually have any experience with systems that run from "consumer
grade flash devices" to back up your doubt, or are you just talking out of
your nether regions? I run dozens of servers with mirrored "enterprise
grade" disks -- because they require either performance, capacity, or write
levels that flash can't handle -- and about a dozen that boot from flash,
because they have none of those requirements. I buy my flash cards from
the shelf at the photo store or at Wal-Mart, and I've never, ever had to
replace one. That includes two cards that have been in continuous service
with _no_ measures taken to reduce write frequency, at a client site to
which I have no physical access, for over five years. I have to replace a
disk in one of the servers with mirrored rotating storage about every six
months on average.
--
Thor Lancelot Simon tls@rek.tjls.com
"We cannot usually in social life pursue a single value or a single moral
aim, untroubled by the need to compromise with others." - H.L.A. Hart
| |
| Anthony J. Albert 2006-06-28, 7:13 pm |
| On 26 Jun 2006 04:45:51 -0700, "Natan" <nvivo.misc@gmail.com> wrote:
>We have some servers like DNS that are very important, and have very
>low disk usage and access. We currently use 2 17Gb SCSI 1 for them, but
>it is hard to find replacement disks and every new disk our people find
>is at least 36Gb.
>
>I think it is a waste to have 36Gb on a server that will never use more
>than 400mb and I was reading that CFs are more durable and reliable
>than hard disks.
>
>We are looking to buy some new supermicro servers and I noticed that
>some 1U servers have IDE CF Card support.
>
>So, my question is: is it a good idea to replace those SCSI RAID 1 with
>simple 1or 2 Gb CF cards? Is it reliable against RAID 1?
>
>And what about logs? Would I need to control how linux writes the log
>or today's CF cards are prepared to this kind of work?
>
>Thanks in advance.
While I haven't done any long-term extensive write-testing, myself, I
have a couple of systems that I operate with CF replacing IDE hard
drives as the primary drive. No problems, given that these systems
are seeing only light use (one Thinkpad notebook, and one 486 running
FreeDOS).
I can verify that there is a measurable speed-up from using CF drives,
under most operating conditions. Most applications load using half or
as little as a tenth of the time needed to load from hard drives, by
my timing experiments.
I would strongly suggest RAID1 for an application as critical as a DNS
server, no matter what media is used to store the data.
Hope this helps,
Anthony Albert
| |
| robertwessel2@yahoo.com 2006-06-28, 7:13 pm |
|
Thor Lancelot Simon wrote:
> In article <1151370106.321739.254800@i40g2000cwc.googlegroups.com>,
> robertwessel2@yahoo.com <robertwessel2@yahoo.com> wrote:
>
> You're a fool, or you've been drinking too much of some fool's kool-aid.
> Semiconductor devices are orders of magnitude more reliable than anything
> with moving parts in it -- and always will be.
>
> The issue with flash is, of course, write cycles. So people whose systems
> boot and run from flash for reliability reasons commonly take simple
> measures to ensure that the flash is not written in normal system operation.
>
> Do you actually have any experience with systems that run from "consumer
> grade flash devices" to back up your doubt, or are you just talking out of
> your nether regions? I run dozens of servers with mirrored "enterprise
> grade" disks -- because they require either performance, capacity, or write
> levels that flash can't handle -- and about a dozen that boot from flash,
> because they have none of those requirements. I buy my flash cards from
> the shelf at the photo store or at Wal-Mart, and I've never, ever had to
> replace one. That includes two cards that have been in continuous service
> with _no_ measures taken to reduce write frequency, at a client site to
> which I have no physical access, for over five years. I have to replace a
> disk in one of the servers with mirrored rotating storage about every six
> months on average.
As I mentioned in my original post, my evidence is anecdotal. In short
my largish collection of dead CF, PC-Card, USB, SD and whatnot flash
devices, plus what I've heard from others on the subject. I've also
not seen any trouble with flash based devices where the flash is
internal, or a vendor-supplied PC-card. Clearly a server will run in a
much more protected environment than the typical consumer flash device,
but the quantity of dead flash devices I've seen doesn't inspire
confidence.
I did a little poking, and while MTBFs for consumer grade devices are
almost impossible to find, the few I did find were in the 250K hour
range (IOW rather less than server class HDs), whereas all the
"industrial" grade devices were 2 million+.
I would expect that a flash device is likely to have a longer useful
life, however. And certainly in harsh environments.
|
|
|
|
|