SSD (Solid State Drive) and pfSense (Important)
-
I read this topic and have doubts. I have SSD OCZ Vertex 4 drive and just installed livecd version of pfSense 2.1
I really need at least 2 packages:- freeradius,
- squid
- vpn
- log everything I can,
- create graphs by Sarg / Lightsquid.
Simple question now - should I change SSD to HDD?
-
I read this topic and have doubts. I have SSD OCZ Vertex 4 drive and just installed livecd version of pfSense 2.1
I really need at least 2 packages:- freeradius,
- squid
- vpn
- log everything I can,
- create graphs by Sarg / Lightsquid.
Simple question now - should I change SSD to HDD?
I would think that in general: no.
Your OCZ Vertex 4 is at least 64GB, unless you're caching for a huge number of people, you shouldn't be wearing out your SSD with pfSense. You have, likely, a huge amount of room for wear leveling, you should be fine. A lot of people were seeing their problems in smaller drives with less robust (or simply not effective) wear leveling.
A Hard Drive will eventually fail; exceptions aside, expect high failure rates after 3 to 5 years, no matter how they're used (exceptions include: defects, high numbers of spin up/down cycles, temperature, vibration, etc.)
Defects aside, an SSD usually lasts until it runs out of cells to swap in for failed cells. The more free space the drive has to easily institute wear leveling, the larger the pool of cells it has to swap with. Many modern drives can also shift data that's been sitting on a cell for a long time and move it to an highly exercised, but still good cell to reclaim more underutilized cells to level with.
Now, you're using an OCZ drive, some of which are arguably more fail prone than some other brands. But, if it hasn't so far, it's probably fine.
-
SSD drives are too expensive when you need only to write some log and RRD informations, use CF cards :
http://www.memorydepot.com/ssd/listcat.asp?catid=icf4000
2 000 000 writes/erases
I just had a hard drive failure a few days ago and the replacement hard drive I had already has been powered on for 5 years. The question is not if it will fail again, but when will it fail? After 5.1 years powered on? 6? Maybe 7? Anyways however you look at it, it's getting towards the end of that bathtub curve. So I want an SSD in there. And since this isn't for a hobby or my home (where I bought an 8GB Kingston SSD that had the WORST possible reviews ever… actually been very reliable, but the model's not sold any longer)
Many systems today no longer have IDE ports. And last time I tried this the SATA adapater was not very reliable. On cold boot I would have to boot the system and then press the reset buton for the CF to be picked up. Otherwise it gets stuck on "No boot device" error. Not very good at all. For this reason I would rather have a single unit from a known/reliable vendor. I was thinking of purchasing something like "Western Digital SiliconDrive A100 8GB MO-297 SATA II SLC Industrial Solid State Drive SSD-S0008SC-7100."
SSD drives are too expensive when you need only to write some log and RRD informations, use CF cards :
http://www.memorydepot.com/ssd/listcat.asp?catid=icf4000
2 000 000 writes/erases
the thing is I use this box also for squid + dansguardian. The performance is very good using SSD. I guess I need to move the squid + dansguardian to another box and replace the SSD with HDD or CF card. 128GB on pfsense without squid is waste of space, better place the SSD on my desktop machine.
I was going to get smallest SSD listed on local store but they only got the 120GB and up :( Regarding industrial CF card you linked, I don't think its available locally here in Indonesia :( I also tried to get sata to CF adapter but I couldn't find it. The closest I could find is USB to CF adapter.
I'm running HDD with Squid + Squidguard. I'm thinking it should run just fine using SSD embedded with a null/no cache.
-
I'd get something larger. That way there is more unused space and the wear will spread out over the rest of he drive (not to be confused with reserved area). If you want long term, get larger even if you don't use it all.
Also SSD's are not 100% tried and true like HDD. It's also about history of the brand/model. -
I've opted for SSD for heat avoidance and silence. This is a house not a real production environment.
Model Family: SandForce Driven SSDs
Device Model: OCZ-AGILITY3
60 gig
Firmware Version: 2.152.0.3-RELEASE (i386)
Intel(R) Core(TM)2 Duo CPU E4600 @ 2.40GHz
4 gigs ram of 4 max possible
intel pro/1000 dual port (EXPI9402PT)firewall, routing, blocking silliness (skype,p2p, et al), hopefully adblocking/pornblocking/malware reduction, hopefully VPN to ip pbx for desktop IP phones without respective vpn, maybe freeswitch for chan_sccp for cisco ip phones more power.
If you want to have VGA and console output at the same time, you can use Hacom nanobsd Pfsense images:
http://www.hacom.net/catalog/pub/pfsense
hooray thanks for this. I am leery of the joy of first time serial install
I'm hoping RRD can be batch stored to an NFSv4 share?
Can hacom/nano/full now use SSD for everything other than logging and store to sata drive?
-
damn.. I just found this thread. I guess I need to replace my 128GB SSD to regular HDD in my pfsense box before it fails.
There is a misconception that SSD:s will fail after using them for a while. Most SSD:s can write an extreme ammount of data before it fails. Compared to a normal hard drive i would think that they probably have alot longer lifespan, especially when in an small Linux/FreeBSD-system like pfsense or another network distribution.
See this thread where you can compare how much data diffrent SSD:s can write. One 64 GB Crucial M4 drive wrote 768 TiB (Thats 0.76 petabyte) before it failed. Lets say you have 100 log entrys every second in your firewall, it would still take years, or decades, to come up to almost one petabyte. And since most SSD:s have a great garbage collection system, if you buy a larger unit (like 128, 256 or 512 GB) then they can write even more.
From a personal view i think SSD:s are great in many aspects. I have a Crucial M4 64 GB in my filserver that has been running for 3+ years, two Intel X25-based that has been running since i bought them, and none of them has failed yet.
In conclusion, don't trust people that tells you that your SSD will fail of too much writing.
-
If you are using 2.1, also see this thread: http://forum.pfsense.org/index.php/topic,63656.0.html
on how to enable TRIM on the filesystem, use the ahci driver (which is required for TRIM), and fix the SMART diagnostic page after changing to ahci. From what I've read, enabling TRIM should help a lot with SSD longevity. -
@Kr^PacMan:
See this thread where you can compare how much data diffrent SSD:s can write. One 64 GB Crucial M4 drive wrote 768 TiB (Thats 0.76 petabyte) before it failed.
…
In conclusion, don't trust people that tells you that your SSD will fail of too much writing.
These guys have been writing to an old 40GB SSD (Intel 320 Series) non-stop for 28 days and failed to make the drive fail…
:D ;D
-
I want to report that I've been running a normal install on a 32GB Intel X25-E SSD since the end of September. I was lucky enough to receive a used drive that was at 99 of its wear indicator on SMART (100 is new). It is still at 99 and I expect this will outlast the pfSense box it is in (considering the free blocks and wear rate). The drive shows 1831 power on hours and 80574 Host_Writes_32MiB.
TRIM is enabled. Logging is enabled. Not a ton of packages are running, but my worries are eased. This is basically one of the first drives to support TRIM (and not fail regularly) and they are very affordable now.
–---
My apologies for resurrecting this thread, but it seemed it would be helpful. It ended with some ambiguity and doubt and I wanted to clear things up. I see later posts that are still confused and I had my own doubts based on what was said here but decided to give it a shot and monitor things.After all, its a bit hard to report on a longer-term test without waiting a bit!
-
You're using a X25-E. The Earth will stop spinning before that wears out.
-
Has anyone tested any mSATA SSDs to see how they handle sudden power loss? There's an Anandtech article that does a better job of explaining the issue I'm concerned about than I could. An excerpt from the "Truth About Micron's Power-Loss Protection" section:
In the MX100 review, I was still under the impression that there was full power-loss protection in the drive, but my impression was wrong. The client-level implementation only guarantees that data-at-rest is protected, meaning that any in-flight data will be lost, including the user data in the DRAM buffer. In other words the M500, M550 and MX100 do not have power-loss protection – what they have is circuitry that protects against corruption of existing data in the case of a power-loss.
So only in-flight data will be lost? I guess I didn't get the memo when they changed the definition of sync. The whole explanation of MLC programming is well worth the read. It pretty much explains the results this guy saw IMO.
I have some of the 16GB PC Engines mSATAs which use a Phison PS3109-S9-J controller. On the Phison site, they reference SmartFlush and GuaranteedFlush trademarks. It sounds like this (PDF warning) which claims they don't ACK a FLUSH CACHE command until data actually exists on (non-volatile) NAND. Based on that, I feel fairly confident using them, but I haven't seen any independent analysis that verifies the claims. Has anyone else? Has anyone used those SSDs? Opinions?
-
3 - 5 years on my Samsung SSDs - 96% - 99% left. No trim. On and writing 24/7 365.
I'm guessing I'm not looking for any failures before I get bore with the drive performance.
-
3 - 5 years on my Samsung SSDs - 96% - 99% left. No trim. On and writing 24/7 365.
I'm guessing I'm not looking for any failures before I get bore with the drive performance.
SLC or MLC on them samsung ssds?
-
SLC…
-
@ryan29:
Has anyone tested any mSATA SSDs to see how they handle sudden power loss? There's an Anandtech article that does a better job of explaining the issue I'm concerned about than I could. An excerpt from the "Truth About Micron's Power-Loss Protection" section:
In the MX100 review, I was still under the impression that there was full power-loss protection in the drive, but my impression was wrong. The client-level implementation only guarantees that data-at-rest is protected, meaning that any in-flight data will be lost, including the user data in the DRAM buffer. In other words the M500, M550 and MX100 do not have power-loss protection – what they have is circuitry that protects against corruption of existing data in the case of a power-loss.
So only in-flight data will be lost? I guess I didn't get the memo when they changed the definition of sync. The whole explanation of MLC programming is well worth the read. It pretty much explains the results this guy saw IMO.
I have some of the 16GB PC Engines mSATAs which use a Phison PS3109-S9-J controller. On the Phison site, they reference SmartFlush and GuaranteedFlush trademarks. It sounds like this (PDF warning) which claims they don't ACK a FLUSH CACHE command until data actually exists on (non-volatile) NAND. Based on that, I feel fairly confident using them, but I haven't seen any independent analysis that verifies the claims. Has anyone else? Has anyone used those SSDs? Opinions?
Sync is only successful if it completes, at which point the data is no longer in-flight. Non-synced data is at the whim the of the write cache.
Samsung has taken the approach of dynamically changing between MLC and SLC and quickly writing data as SLC, the re-writting the data as MLC as time permits. This allows sync to return quickly.
-
3 - 5 years on my Samsung SSDs - 96% - 99% left. No trim. On and writing 24/7 365.
I'm guessing I'm not looking for any failures before I get bore with the drive performance.
SLC or MLC on them samsung ssds?
The new 850 use vNAND which can be tweaked between performance and longevity but with no sacrifice to storage density. vNAND can have between 3,000 and 40,000 write cycles, depending on how they tune it. The difference in performance is only between 35MB/s and 55MB/s per chip. I would gladly take 35MB/s for a 10x improvement in write cycles.
One tech site has their 850 pro up to 1.5PB written and it still has 60% of the reserve pool left. It has been at 0 wear leveling for months. Samsung claims they have an 850 120GB drive with 8PB written and it still works.
5 year 170TB written warranty or 10 year for the pro, but same written.
Samsung said they have accepted warranty claims with drives over the 150TB limit, assuming they were using non-server loads, but I wouldn't bank on it.
-
Has anyone used a SATA DOM's for an installation? I've got one in my freenas box and I'm real happy with it.
Would a 16GB SATA DOM suffice or should something larger be used for wear leveling?
I plan on updating hardware at some point and was thinking of using one for a small footprint silent setup unless they aren't recommended.
-
Some of the SATA SSD DOMs (disk on module) sound interesting but unless space is really at a premium I think a regular SSD is going to be better priced. Really dig into the specs if you plan on reading/writing a lot too, I'd avoid any that don't have a full function controller in any case.
http://www.thessdreview.com/our-reviews/mach-xtreme-sata-dom-32gb-ssd-review-small-os-storage-postage-stamp-sized-form-factor/
-
SLC…
at the moment, i believe that SLC SSDs are becoming more difficult to find nowadays than its MLC counterpart. i wonder, however, if by using enterprise SSD would suffice over consumer SSD. e.g., the samsung 845DC EVO over the 840 pro. what else or other options is there for us who are about to get ready to build our own pfsense router?
-
I would look for an MLC SSD with a year or two of ratings not showing any issues if price is a concern. Lots of people are reporting intel to be very solid performer.
I don't trust ANY super new tech in a pfsense install I have to rely on to just-work but if its got a bit of time on it and good ratings, should be ok.