SSD Raid1 vs 15k Raid10

[ibaker]

Can someone tell me which of the following server hdd configuration would provide the best performance for XF with say 200 concurrent users, 250k posts, 30k threads times 3 sites (hoping to end up with 3 sites)
- 2 x 300gb Intel SSD hdd Raid 1
- 4 x 73gb 15k rpm sas hdd, Raid 10

 

[Slavik]

It depends on the spec of the drives.

What models are they? Generally though, the ssd's will be better.

 

[ibaker]

Unfortunately that is all I know at this stage...I have a choice of either...currently I am using the SAS Raid10 configuration and need to know if it is worthwhile going to the SSD

 

[tyteen4a03]

One is 300GB while the other is 146GB.

Would everything (including attachments, images, cache data) be on the same drive?

It depends on the spec of the drives.


What models are they? Generally though, the ssd's will be better.

I heard that Intel SSDs are the best ones in the market.

 

[kontrabass]

Go with the Intel SSD's! I used to have 15k drives in RAID 10 on my database server for years. On my new server, I have Intel SSD's in RAID 1 that smoke the 15k drives. Granted the processor on the new server is superior as well - but it's pretty clear the bottleneck was disk IO for us.

 

[CTXMedia]

The bottleneck in my old server was the HDD - so for my latest server I went with an Intel 520 240GB SSD - I didn't bother with RAID, instead I just sync to second Intel 520 SSD every four hours with a cron job, as well as sync to a regular HDD overnight, and also to Amazon S3 weekly.

I overprovisioned my SSDs by 30% to add a little extra longevity.

I didn't go for RAID as the controller would add another point of failure to the box, and SSDs in RAID1 seem a bit pointless as you'll wear out the write cycles on each drive equally. Sure, you get some failover protection, but the money I saved went towards bumping up the RAM - enabling more data to be cached and kept in memory.

I moved everything over (see CycleChat link in sig.) early January and after some tweaking and fine-tuning it's been running sweet as a nut since (with a very low load and very fast response).

Unfortunately I don't have a comparison with SAS but am happy with the choice I made and so are my members.

Cheers,
Shaun

 

[Luke F]

I heard that Intel SSDs are the best ones in the market.

Old intel SSDs are good, but newer ones (Intel 520 series) are some of the worst you can buy due to Sandforce controllers being used in them. Pretty much guaranteed to fail in typical HDD lifetime if not less (disregarding flash wear), when an SSD should last very significantly longer (again disregarding flash wear), and of course the low speeds with incompressible data (aka the vast majority of data).

For a server, Samsung SSDs (830 and 840 pro) are where it's at. Crucial are good too, but a poor reputation with firmware bugs.

In general any manufacturer that has a large say in as many as possible of: -the controller, -the flash chips, -the firmware, will produce good SSDs. If I'm not mistaken, Samsung is the only one with all 3.

 

[ibaker]

I was thinking of Raid1 to protect from drive failure as the SSD stories that are around say they have a lot shorter life span...I will also have a 7,200rpm SATA disk for daily backups.

The full story is I am trying to decide if an upgrade of my dedicated server would be beneficial...my data centre let me down big time recently and whilst they have given me a months hosting for free I know thay also have another box that I am wondering whether I should fight to get as restitution however I need to know if it would be worth the fight as in I would get far better performance than what I am currently getting.

My current box is:
Intel e3 1230 (8 cores), 16gb ram, 4x73gb 15k rpm sas hdd, raid 10, 320gb sata II hdd, 10TB, 100mbps port

The other box is:
Intel i7 3930k super fast 6/12 core 3.2ghz, 16gb ram, 2x300gb Intel SSD hdd raid 1, sata II hdd, 10TB, 100mbps port

If I fight really hard I may get the other box for the same price as I am paying now for my current box BUT I may end up having to pay a little more so I need to know whether the fight would be worth it i.e. not worth fighting for the other box if I am not going to get remarkably better performance but instead receive free hosting on my current box for x months

 

[WSWD]

The issue between the two boxes is that one is server-grade equipment and the other is desktop-grade equipment. Yes, the 3930 is going to be fast, but it's a desktop system, with non-ECC RAM, etc. Is the 3930 going to have hot-swap drives, redundant power supplies, etc.? Who knows, but it's worth questioning. Most reliable datacenters won't touch desktop-grade equipment with a 10 foot pole. We use the E5-1650 (basically the equivalent to the 3930) on our main website, and it is definitely a quick CPU. It will run circles around your 1230.

If you are potentially going/willing to pay a little more, it would be worth asking what your current box specs would cost with 4 x Samsung 840 pros in RAID 10. That would be a massive improvement over the SAS array.

 

[Ghan_04]

Old intel SSDs are good, but newer ones (Intel 520 series) are some of the worst you can buy due to Sandforce controllers being used in them. Pretty much guaranteed to fail in typical HDD lifetime if not less (disregarding flash wear), when an SSD should last very significantly longer (again disregarding flash wear), and of course the low speeds with incompressible data (aka the vast majority of data).

On what evidence are you basing that conclusion?

http://www.anandtech.com/show/5508/...cherryville-brings-reliability-to-sandforce/9

"The 320 is actually widely used in servers as it's very reliable and can last a good amount of time with the right amount of over-provisioning. The 520 just destroys it. The bigger benefit is that if you're dealing with a workload that's not already compressed, the 520 will guarantee you much better drive longevity than the 320 thanks to the fact that it's simply not writing as much data to NAND."


http://www.storagereview.com/intel_ssd_520_review

"To mitigate reliability concerns Intel has invested substantially to make sure the 520 lives up to user expectations. This means the SSD 520 has been validated in 120 client system types and has passed over 5,000 unique tests during the validation cycle. These tests include things like: electrical validation, signal integrity, unplanned power loss, SATA error injection, temperature monitoring, workload testing and so on. To further ensure consumers feel safe buying the SSD 520, Intel is including a five year warranty, which effectively protects customers for the projected useful life of the drive."


http://www.tomshardware.com/reviews/ssd-520-sandforce-review-benchmark,3124-15.html

"Intel's big opportunity is combining familiar second-gen SandForce controller performance with typical Intel reliability. It joins OWC as one of the only companies offering such long-term warranty protection on prosumer-oriented SF-2281-based SSDs.
But the highest bin of flash memory costs more, as does guaranteeing a piece of hardware for two or three years longer than the competition. Other vendors rely on SandForce's RAISE technology to maintain data integrity, even as they use NAND world's equivalent of well liquor to mix up your SSD. And they're able to push prices down as a result.
The question then becomes: would you rather pay a premium for high-quality NAND, more confidence-inspiring warranty coverage, and Intel's support structure, or do you have enough faith in SandForce's controller technology to compensate for the use of lesser memory at a lower price point?"


http://www.behardware.com/articles/881-7/components-returns-rates-7.html


From everything I've read, Intel is one of the most reliable SSD manufacturers out there. I would have no problem recommending Intel SSDs for a server box.

 

[ibaker]

My current box already has massive headroom compared to what I need but at $179 a month it is worth it for peace of mind, opportunity to expand and spare resources for attacks etc. The "other" box is only $149 a month but only has a single SSD whereas I want some redundancy so the extra SSD disk and SATA backup disk would normally cost an extra:
300gb Intel 320 - $45
Raid 0/1 - $20
500gb Sata II 7200 rpm - $10

I am fighting to get that included without paying anything more than the $179 I am already paying but this thread was about whether I should fight for that if it isn't really going to give me any greater performance increase over what I have now...NOTE: I am NOT a server admin person and know very little about servers.

The issue started back in Jan when they changed/updated a file on my server without telling me so I could first advise my server admin people for comment and the server crashed big time...they said they had done nothing yet after 12hrs of it being down we finally ascertained that they "had" done something.

Since then the server has never performed properly with issue after issue...my server admin people were unable to resolve several major intermittent issues. I suspect that these were caused with all the tooing and froing trying to get my server back up again.

Last weekend I had enough and decided to start the box again, fresh, so backed up my sites to the 5th disk, the backup disk, and paid them to do an OS restore. They were advised several times over several days that ALL my data would be backed up on the backup SATA disk. Just prior to the OS restore they asked me to confirm that I had backed up the data to the backup disk. After doing the OS restore they advised that in the process they also wiped the backup disk.

I was paying a server admin guy to set the box back up with MySQL, PHP etc...after he had done that he also then advised that they had installed a very old version of the OS as well...aaargh

One lucky thing was that I had just prior to the OS reload also backed up everything and downloaded it to my PC (disaster recovery and this was a disaster) so I then had to upload the data to the box in LA from here in Melbourne Australia at a painstakingly slow rate of between 12k/sec to 75k/sec taking forever. Not only that but using MySQLDumper to import a 800MB db into mySQL took hours as well. 24hrs later my server is back on air but now running old software versions.

One consolation is that the OS restore has fixed my issues that I was getting after what they did back in Jan

Overall I am not a happy camper as I now have to go through it all again to get my box back to recent software and recent OS. Going to another box would give minimal impact so the option of going to the SSD box is appealing but only if I will also get far better performance...hence why I started this thread with my question

 

[p4guru]

ibaker, bit confused are your web host and server admin, the same company ? who actually changed/updated a file (which file? ) that caused the server crash ? the web host or server admin ?

I ask, because if they aren't the same company and the problem was with the server admin, then there shouldn't be any dramas in keeping the web host and just dumping the server admin ? I say such as the hardware specs for the price are very competitive on either server specs you listed.

As to choice between drive configs, the 2x 300GB Intel SSD raid 1 will be way faster - for that capacity i'd assume it's the Intel 300GB 320 series SSDs. Great reliable drives, I have clients running these Intel 320 series SSDs in 4x and 8x disk raid 10 setups for nearly a whole year without problems.

But benefits of SSD will only help if you're disk I/O bottlenecked.. if the slow performance bottleneck is elsewhere, it won't help at all.

 

[ibaker]

Thanks Mate...I have 3 entities that I use:
1. The Data Centre I use WebNX and they are the ones that changed the file back in Jan without informing me and who supply and maintain the hardware...they also do the initial OS setup and install WHM/cPanel
2. I then have a Server Admin person that I know and respect, now a friend, who is the Ops Mgr at a high end data centre here in Melbourne during the day...great for doing initial setups outside of his normal working hours and charges what he is definitely worth
3. To monitor and maintain the server on a daily basis I use a Server Admin company they monitor it with continual pinging, do general maintenance like adding PHP modules, maintaining mySQL, installing 3rd party software (within reason), and fix any issues if they happen etc...They are an Indian company (here) but I have lost faith in them given the issues over the last couple of months, I use to use this company but they take a long time to address any problems and then deny it...I know both cheap but I can't afford better

I wouldn't have a clue where any bottlenecks are...all I know is that currently the server is in LA and the majority of my site users are in Australia so the faster I can get the data coming out of the box will minimalise the perceived delay of latency to the users and I hope to soon have more sites catering to users in other countries so need room to grow without having to change servers...for all I know the current server is ok but if the disks and processors can get the data going out faster that surely must benefit the users...perhaps I have it all wrong but as mentioned, I know zilch about servers

I really want to say a big thanks for your help on this guys...my site is about aviation and I can tell you the average life expectancy of a VFR pilot in IMC is 178sec BUT if smoke is coming out of a server is there something wrong or just a habit it needs to kick

 

[CTXMedia]

If the majority of your users are in Australia - why not look for a good host in Oz?

It'll be the best way of reducing latency. SSDs aren't going to fix transatlantic data transmission delays.

Then look at using a CDN for your worldwide audience.

 

[Luke F]

On what evidence are you basing that conclusion?

http://www.anandtech.com/show/5508/...cherryville-brings-reliability-to-sandforce/9

"The 320 is actually widely used in servers as it's very reliable and can last a good amount of time with the right amount of over-provisioning. The 520 just destroys it. The bigger benefit is that if you're dealing with a workload that's not already compressed, the 520 will guarantee you much better drive longevity than the 320 thanks to the fact that it's simply not writing as much data to NAND."


http://www.storagereview.com/intel_ssd_520_review

"To mitigate reliability concerns Intel has invested substantially to make sure the 520 lives up to user expectations. This means the SSD 520 has been validated in 120 client system types and has passed over 5,000 unique tests during the validation cycle. These tests include things like: electrical validation, signal integrity, unplanned power loss, SATA error injection, temperature monitoring, workload testing and so on. To further ensure consumers feel safe buying the SSD 520, Intel is including a five year warranty, which effectively protects customers for the projected useful life of the drive."


http://www.tomshardware.com/reviews/ssd-520-sandforce-review-benchmark,3124-15.html

"Intel's big opportunity is combining familiar second-gen SandForce controller performance with typical Intel reliability. It joins OWC as one of the only companies offering such long-term warranty protection on prosumer-oriented SF-2281-based SSDs.
But the highest bin of flash memory costs more, as does guaranteeing a piece of hardware for two or three years longer than the competition. Other vendors rely on SandForce's RAISE technology to maintain data integrity, even as they use NAND world's equivalent of well liquor to mix up your SSD. And they're able to push prices down as a result.
The question then becomes: would you rather pay a premium for high-quality NAND, more confidence-inspiring warranty coverage, and Intel's support structure, or do you have enough faith in SandForce's controller technology to compensate for the use of lesser memory at a lower price point?"


http://www.behardware.com/articles/881-7/components-returns-rates-7.html


From everything I've read, Intel is one of the most reliable SSD manufacturers out there. I would have no problem recommending Intel SSDs for a server box.

Sandforce failures are all over the place

eg http://communities.intel.com/thread/33953

I don't doubt that intel can write a great reliable firmware, or that they can source high binned NAND chips. But when they've got a Sandforce controller in there, none of that matters.

 

[Forsaken]

Sandforce failures are all over the place

eg http://communities.intel.com/thread/33953

I don't doubt that intel can write a great reliable firmware, or that they can source high binned NAND chips. But when they've got a Sandforce controller in there, none of that matters.

Don't forget about the AES256 issues as well. They marketed these as being some of the best for encryption and security yet couldn't use AES256 (Not sure if this has been rectified).

Regarding the OP: if peace of mind is important to you, I would recommend looking at a new host. The fact that they did something without your knowledge or your heads up and then tried to pretend they didn't shows a lack of professionalism. A free month of service or a server upgrade is all nice and good, but having a host that you can trust is much more important.

If it were me, I would dump them and look for a new provider in the free month I have from them. Generally if you contact a sales department of a hosting company and explain your situation and needs you can often get better prices than advertised .

 

[WSWD]

Wow! Scary to hear that you are using WebNX, and they wiped the backup drive too. That's our major west coast provider for our servers. We have never had an issue with them, and have been with them for 6+ years. If you aren't happy with WebNX, and want to stay in CA still, you can also give Quadranet a try.

The SSD drives are going to be faster, but the latency, as was already mentioned, is coming from the fact that the server is located in CA and your users are in Australia. That is definitely going to be the quickest path from the US to Australia, but if you really want to speed things up, you should get a server in Australia. That will cut down on the latency more than anything else you will do to that server.

 

[Floren]

I would go with the physical drives, not SSD. While being a lot faster, SSD's have a limited life time, based on the writes you perform on them. In Linux, you can check the degradation factor easy:

# smartctl -a /dev/ada0

But then, you have to monitor their state on a monthly basis. On the other hand, 4 15k disks will still offer you amazing performance for your site.

 

[ibaker]

As far as the latency goes, yes I know Aust hosting would be far quicker but 2 things, the cost of hosting here in Auistralia is 3 times that of the US...I would be lucky to get a decent 2gig ram VPS for that money, and secondly I want to expand with other sites specific to other countries. A West Coast US server provides a happy medium across everything

 

[Slavik]

I would go with the physical drives, not SSD. While being a lot faster, SSD's have a limited life time, based on the writes you perform on them. In Linux, you can check the degradation factor easy. But then, you have to monitor their state on a monthly basis. On the other hand, 4 15k disks will still offer you amazing performance for your site.

Monthly lol.

I might check mine once every 6 months if that.

On an average forum, where the real data throughput on the disk in real terms is absolutely minimal, SSD's will probably last *YEARS* before any issues arise.