How many of you really need a Pure SSD VPS?

Nekki

wych said: Now your a mod can we hold a LET-BenchOff?

Sure, do whatever you want.

wych

@Nekki said:

I present, The BenchOff.

Hosts - Don't hate me.

alexh

@GelHost said:
People prefer something because they hear someone saying "o ya that is the best". Very few people have the tech skills to know if this is really true or not.

I don't think customers lose from being on an SSD-based node. Solid-state drives have virtually no random access time; This usually directly improves SQL and web server performance (as an example).

tdc_adm

Corey said: Sounds like you need to optimize your code and indexes

Yes, I should. But the result will not improve very much because our site have many different queries with different criteria. Then the SSD comes

eddynetweb

@GelHost said:
Anyone uses SSD for their website?

I do, and it works great. MySQL also works faster.

GelHost

@eddynetweb said:
I do, and it works great. MySQL also works faster.

How do you know it work faster, have you done some testing?

AThomasHowe

GelHost said: How do you know it work faster, have you done some testing?

Are you proposing MySQL (or anything really) runs faster on a HDD over an SSD?

GoodHosting

@AThomasHowe said:

I submit that anything run on a single Constallation ES.5 1TB HDD will run rather similarily on a single Intel "Datacenter" SSD (the crap that SingleHOP and a lot of other "Enterprise" companies call an SSD.)

Please note the following piss poor statistics from the Intel "Datacenter" SSDs:

Sequential Read            340 MB/s
Sequential Write           100 MB/s

Random Read (100% Span)  70000 IOPS
Random Write (100% Span)  7000 IOPS

Latency - Read              50 µs
Latency - Write             65 µs

Note the absolutely terrible HDD-like sequential writes, and the terrible IOPs capability of these "SSD" drives that Intel shits out? Now, let's compare a real performance SSD below (Samsung 840 PRO 256GB):

Sequential Read            540 MB/s
Sequential Write           450 MB/s
Random Read (100% Span)   100k IOPS
Random Write (100% Span)   78k IOPS
Latency - Read              40 µs
Latency - Write             55 µs

It's actually rather appalling how terrible the Intel drives are.

GoodHosting

Addendum:

Here are some fun facts about the actual bottlenecks you're going to face:

SATA 3 (6GB/sec) can only (IN LAB CONDITIONS) be saturated at 600MB/sec of line speed. You cannot significantly surpass this limit unless you are not using the SATA interface at all, ex: PCI SSD.

** The VPS providers that don't just use Software RAID [ and trust me, there are loads of these... ] ; often use inferior Hardware RAID cards that are not rated for SSDs due to their low or nonexistent onboard cache, and low-clocked or MISSING onboard processor, ex: **

• LSI MegaRAID 9240 : NO ONBOARD CACHE, NO ONBOARD PROCESSOR
• LSI MegaRAID 9260 : 512MB ONBOARD CACHE

The above cards are NOT RATED for SSDs.

AThomasHowe

I guess you're right @GoodHosting but how many providers are going to be using such low r/w drives? I could just as easily compare it to the worst HDDs and make those Intels look good. There's always gonna be bunk products on the market.

I guess my statement did kind of imply ever.

GoodHosting

@AThomasHowe said:
I guess you're right GoodHosting but how many providers are going to be using such low r/w drives? I could just as easily compare it to the worst HDDs and make those Intels look good. There's always gonna be bunk products on the market.

I guess my statement did kind of imply ever.

SingleHOP uses these drives exclusively. I found a few kicking around other providers as well, and it's rather appalling to me. Their excuse is the fact that the Intel drives have a 20% longer MTBF, but with Hardware RAID10 over 8 drives I'm not all that concerned about 250,000 hours worth.

rootwelt

You will never use HDD's again once you had the expericense with SSD storages. Maybe for some kind of file garbage HDD's are OK, but the server core on SSD looks much better.

Microlinux

K2Bytes said: Outside LET, the vast majority of mature users still prefer SATA over SSD or SAS.

The only real reason to prefer a spinning disk over an SSD would be if you need a lot of space. I'm not sure why a sane person would purposely choose the slower, less reliable media if they didn't need that.

It's more likely the vast majority of users outside LET that you claim "prefer" spinning disks (please quote your source of this information) are simply with hosts who are still running them or need the space. They aren't going to leave for a provider who uses SSD if their existing service is working fine for it's purpose.

Maounique

There are cases of big servers or SANs where the number of disks and huge cache int he controllers make up for the SSD advantages in terms of IOPS. As for sequential WRITE, an enterprise SATA disk is going to be close to an enterprise ssd, the difference is made in controller cache and in random seek time which leads to way better iops drive per drive, but not so much so with big caches and 24+ disks.
4 disks per server and lowly raid controller or software? Use SSD in raid 10. 12+ disks and 2 GB cache? Definitely go with 4 TB sata disks in raid 5-6. Will give plenty of storage AND low io wait. 24 disks is even better.

GoodHosting

Isn't it obvious?

PCI SSD = VERY Expensive, Small-ish

SSD = Expensive, Small-ish

SAS = Expensive, Rather Small

SATA = Cheap and Huge

Is it rocket science to some of you?

Microlinux

@GoodHosting said:
I submit that anything run on a single Constallation ES.5 1TB HDD will run rather similarily on a single Intel "Datacenter" SSD

Uh, no.

Many things could, certainly not "anything".

Maounique

Microlinux said: Many things could, certainly not "anything".

Yeah, for a busy database with small records and many non-repeating queries from multiple tables impossible to cache in memory for various reasons, even the lowest ssd will do better than a sata disk, no matter how good. But, as you can see, the circumstances must be a bit too specific...

alexh

@GoodHosting said:
It's actually rather appalling how terrible the Intel drives are.

Intel doesn't just produce one line of drives. S3700 200GB had much better results for me. Nowadays you could probably even use 730-series drives, despite them not being enterprise grade.

Also, comparing anything Seagate to anything Intel is appalling; Myself and many members of NCIX vowed to never buy Seagate again, opting for Western Digital instead. I've had many, many Seagate drives fail on me: both desktop and CS.

Microlinux

@Maounique said:
the circumstances must be a bit too specific...

In some sense, yes. But there are most certainly situations outside of what you described where an SSD would be far superior . . . situations where random IO performance is paramount.

eddynetweb

@GelHost said:
How do you know it work faster, have you done some testing?

I just noticed a speed increase in applications and scripts that utilize MySQL.

Shoaib_A

@Microlinux said:
It's more likely the vast majority of users outside LET that you claim "prefer" spinning disks (please quote your source of this information) are simply with hosts who are still running them or need the space. They aren't going to leave for a provider who uses SSD if their existing service is working fine for it's purpose.

Yes that is because if you're with a good host who neither oversells nor fools the customers by manipulating the results of disk I/O with cache controller memory you're not going to see a huge difference in performance.Pity is those who use HDDs put 50 to 100 sometimes even more more customers per node which affects the performance.

I have been running & managing hundreds of database intensive game servers, a forum & a few websites for a last few years both on HDDs & SSDs on leased dedicated servers virtualized with KVM. As profit is not an aim here(all these game servers for the community are funded by me & a few other friends) I distribute the load efficiently across nodes & there is no difference in performance. SSDs for example on RAID 1 can handle more number of intense databases & I/O intensive applications but if you put loss on HDDs(less number of vps/disk intensive databases/applications) would mean superb performance.

And you must be a sharp observer of things going around in the industry to know what people prefer, there is no need of quote tot he source or anything.

Microlinux

@K2Bytes said:
I distribute the load efficiently across nodes & there is no difference in performance.

Because you haven't run a single application (or group of applications) that exceed(s) the performance limits of spinning disks.

Microlinux

@K2Bytes said:
And you must be a sharp observer of things going around in the industry to know what people prefer, there is no need of quote tot he source or anything.

Please find me one credible person who would prefer a less reliable, slower and more power hungry spinning disk over an SSD, if space is not a factor.

alexh

@K2Bytes said:
And you must be a sharp observer of things going around in the industry to know what people prefer, there is no need of quote tot he source or anything.

IT professionals have varying opinions: they're often based on success with a specific task in the past, as opposed to finding the most efficient way of doing said task. (i.e. If it isn't broken, don't fix it)

It's common sense that one should use the type of disk that fits the application well. You wouldn't use a RAID-10 array of SSDs in a backup server because the I/O wouldn't be utilized and it wouldn't be cost-effective.

However, I'd assume that LEB users often combine services like HTTPD and MySQL in a single VM; In this case, solid-state drives or caching would be worth it in my opinion because these services can both benefit from improved random I/O.

Shoaib_A

@Microlinux said:
Because you haven't run a single application (or group of applications) that exceed(s) the performance limits of spinning disks.

Run any call of duty series servers with 1000+ slots filled 24/7 with 50 big brother bots filled & you will see what is more intensive

Shoaib_A

@Microlinux said:
Please find me one credible person who would prefer a less reliable, slower and more power hungry spinning disk over an SSD, if space is not a factor.

There are lot of people who understand very well if they need SSDs or not, you just need to notice the world of hosting industry which exists outside LET.

Shoaib_A

@alexh said:
However, I'd assume that LEB users often combine services like HTTPD and MySQL in a single VM; In this case, solid-state drives or caching would be worth it in my opinion because these services can both benefit from improved random I/O.

Would still be fine if there are not more than 15-20 vps per node but that would be impossible to do within LEB price range.

Microlinux

@K2Bytes said:
Run any call of duty series servers with 1000+ slots filled 24/7 with 50 big brother bots filled & you will see what is more intensive

What on earth does that have to do with anything? Obviously, if your application works fine on a spinning disks, you are not exceeding the disks' limits.

Here's something wild . . . just because you haven't personally run into situations that exceed the limits of spinning disks, doesn't mean they don't exist.

I don't know much more basic I can get, so that's all I have.

alexh

@K2Bytes said:
There are lot of people who understand very well if they need SSDs or not, you just need to notice the world of hosting industry which exists outside LET.

I was under the impression that the entire hosting market was interested in SSDs.

Edit: Added big red circles for clarity.

Microlinux

@Microlinux said:
Please find me one credible person who would prefer a less reliable, slower and more power hungry spinning disk over an SSD, if space is not a factor.

@K2Bytes said:
There are lot of people who understand very well if they need SSDs or not, you just need to notice the world of hosting industry which exists outside LET.

Huh? Did you quote the wrong post?