Does anyone know the cache buffer size on a MARSHAL 4TB SAS HDD

[skippy258]

Hi all I can not find much in the way of specs for a MARSHAL HDD that i'm looking at, but wish to know the cache buffer size on a MARSHAL 4TB SAS 29 Pin HDD

Also how come all SATA 4TB HDD's only 5600 rpm and SAS 7200 rpm when all you have to do too get the extra speed is buy a SATA to SAS 29 Pin?

 

[OneMoar]

generally server drives are at least 64MB
and no use a adaptor will not get you any extra speed
the speed of the link has nothing todo with the drive speed
WD makes a few 4TB 7200RPM drives
but seriously if long term reliable storage is your goal then a single drive is not the way to go you should be using RAID

 

[skippy258]

generally server drives are at least 64MB
and no use a adaptor will not get you any extra speed
the speed of the link has nothing todo with the drive speed

Thanks for your reply, I found a photo of one online. The speed of the drive SAS speed is 6GB same as SATA but the disk spin up is faster that's what I like

 

[OneMoar]

most platter drives are going to top out at 100MB/s sustained r/w
that drive might manage 80 under heavy load

 

[skippy258]

most platter drives are going to top out at 100MB/s sustained r/w
that drive might manage 80 under heavy load

Sweet 80 sounds fine, I will get one drive first and test it out, if I like it.... I might buy another and RAID them on this system.

 

[newtekie1]

Also how come all SATA 4TB HDD's only 5600 rpm and SAS 7200 rpm when all you have to do too get the extra speed is buy a SATA to SAS 29 Pin?

There are plenty of 7200RPM 4TB drives.

7200RPM and 128MB Cache

 

[kn00tcn]

who is marshal? is it still WD or maxtor on the inside?

most platter drives are going to top out at 100MB/s sustained r/w
that drive might manage 80 under heavy load

wat

not my screenshot but similar to one i've used

the more density, the less need for rpm

 

[OneMoar]

note I said sustained R/W meaning simultaneous read and write
and your graph shows that it drops like a stone the longer the operation runs

 

[Aquinus]

note I said sustained R/W meaning simultaneous read and write
and your graph shows that it drops like a stone the longer the operation runs

It's where it's reading from the platter and this can be attributed by how quickly different parts of the platter are going past the head depending on the position from the center. As a result, the outside of the platter has a higher linear velocity than any point on the inside whereas their angular velocity is the same. In summary, all this means is the reads/writes are fastest on the outside of the platter because of linear velocity whereas the inside has the slowest. Performance on platter drives varies based on where on the platter you're reading/write to/from.

 

[OneMoar]

It's where it's reading from the platter and this can be attributed by how quickly different parts of the platter are going past the head depending on the position from the center. As a result, the outside of the platter has a higher velocity than any point on the inside whereas their angular velocity is the same. In summary, all this means is the reads/writes are fastest on the outside of the platter because of linear velocity whereas the inside has the slowest. Performance on platter drives varies based on where the on the platter you're reading/write to/from, nothing more, nothing less.

thats assuming that the drive is writing from one end of track to the other which isn't a real-world test case by any means you have fragmentation,overhead,other i/o querys to be dealt with
its why I never bother benching drives to many variables for the result to ever be consistent enough

 

[Aquinus]

assuming that the drive is writing from one end of track to the other which isn't a real-world test case by any means

Right but, I do think that is how HDTune works. Random reads/write are a whole other story because now you're talking about seek latency and how NCQ impacts access latency and performance. I think we can easily say though that all platter drives blow at random reads/writes. The smaller the platter is (width wise,) and faster the sucker spins, the better randoms are. So those 2.5" 15k RPM SAS drives tend to do pretty good on randoms for being platter drives.

 

[OneMoar]

Right but, I do think that is how HDTune works. Random reads/write are a whole other story because now you're talking about seek latency and how NCQ impacts latency and performance. I think we can easily say though that all platter drives blow at random reads/writes. The smaller the platter is (width wise,) and faster the sucker spins, the better randoms are. So those 2.5" 15k RPM SAS drives tend to do pretty good on randoms for being platter drives.

thats exactly how HDtune works its why I don't put any stock in it
iirc hdtune only runs its write benchmark if the disk is raw and unpartitioned

 

[Aquinus]

thats exactly how HDtune works its why I don't put any stock in it
iirc hdtune only runs its write benchmark if the disk is raw and unpartitioned

What about ATTO? This is from a few months ago on my RAID-5. No benchmark will capture every use case though. I guess this still probably isn't doing randoms though.

 

[newtekie1]

the more density, the less need for rpm

Not true. RPM affects latency more than raw read/write speed. A higher RPM drive will give better performance when reading/writing multiple files or on a highly fragmented drive.

note I said sustained R/W meaning simultaneous read and write
and your graph shows that it drops like a stone the longer the operation runs

Sustained R/W means reading one constant big file. It does not mean simultaneous reading and writing.

The graph drops because reading from the edge of the disk is faster than reading from the center, because the outside of the disk is moving past the read head faster than the inside. HDtune just reads from the edge and goes to the center in one long read. It has nothing to do with how long it is reading, and everything to do with where it is reading on the disk.

thats assuming that the drive is writing from one end of track to the other which isn't a real-world test case by any means you have fragmentation,overhead,other i/o querys to be dealt with
its why I never bother benching drives to many variables for the result to ever be consistent enough

That is how HDTune tests. And why there are other disk benchmarks that give other usage scenarios. HDTune simulates reading one large file from the drive. There definitely are times where that usage scenario applies.

The point is there definitely are drives that will give more than 100MB/s in certain use cases. I often get over 100MB/s when transferring large files from my SSD to my storage drive.

 

[95Viper]

who is marshal? is it still WD or maxtor on the inside?

It looks to be... a Company in Japan that sells products produced by Toshiba; however, they are branded Marshal...
The web page for [HDD discount market] shop.marshal-no1.jp
MARSHAL Corporation Hashizume Toshinari
Yubinbango101-0021, Chiyoda-ku, Tokyo Sotokanda
Three Sotokanda building
03-6803-0106

 

[kn00tcn]

note I said sustained R/W meaning simultaneous read and write
and your graph shows that it drops like a stone the longer the operation runs

100mbytes/s is not a stone O_O, it's above the number you stated, that's my only point (EDIT: sustained as newtekie stated is usually the word used for sequential, that's why i knee-jerk'ed at the speed)

hdtune also has file benchmarks like the other tools

anyway the speed depends on the usage (what kind of scenario reads AND writes for a sustained period other than extracing an archive to the same physical drive? or i guess some kind of multi user server... after all these years of dropping SSD prices or ram caching)

i just want to maximize the speed within the limitations of mechanical, for example choosing a drive with fewer platters at the same total size results in more performance, it's even possible for a single platter 5400rpm drive to be faster than a 7200rpm multiplatter drive