1

Seagate Archive 8TB Benchmark and review

Seagate Archive 8Tb Benchmark

We have been testing and benchmarking the Seagate’s new 8TB Archive HDD, which features SMR, Shingled Magnetic Recording. Described in the following video:

It is marketed as enterprise nearline storage – with the accompanying 5 year warranty! This is a highly power efficient drive with just 5W idle power consumption, average of 7.5W operating and 1W standby. Featuring 128MB Cache in 5900RPM spindle speed.

Being very cost-efficient and 5year warranty means we could potentially utilize these for 6 years. Our standard operating procedure is to replace disks for new ones after their warranty period expires or move them to single user and lower cost services with RAID1. We prefer the former as replacing failed disks cause costs and downtime.

Considering our Dragon series of seedboxes, the largest is 4000GiB, 4 of these drives in RAID10 provides about 15 000GiB of usable storage space, we could put 3x Temeraire and 1xToruk on a single node – It’s important to take a look at these drives!

Seagate Archive 8Tb Benchmark

We did this from our perspective, since we are going to use them for seedbox servers, those are the figures we tested. We did the most important tests for multiple times and chose the median figure, and sometimes averaged or took the single result (SSD). We pitted these drives against several other servers, the most important probably being our own datacenter with 4x3Tb RAID5 setup which shares exactly the same hardware for other than the disks. We use the Seagate’s 3Tb Desktop HDD aka Barracuda, ST3000DM001 and Toshiba 3Tb DT01ACA300 drives in these.

The other servers we pitted against was a 3rd party datacenter servers with unknown discs and hardware raid 0. No direct access to drives. The models were 2x2Tb 7200RPM and 1x160Gb SSD just for comparison. We took two of the 2x2Tb servers and took better median of their results, to get as accurate results as possible. There’s quite a bit of variance with these 2x2Tb HW Raid0 servers so it’s better to take the better one.

For the IOPS tests we used the excellent and simple Benjamin Schweizer’s IOPS tool. The relevant thread counts are 64 and 128 for us.

The tests stops after the IOPS is below thread count by default – However we needed to change this as some of the benchmark servers were not capable of sufficient IOPS to test at relevant block sizes of 512KiB and 1024KiB at high thread counts! We changed this to as long as IOPS is above 16 and blocksize max 16MB.
You can download the modified code from: http://pulsedmedia.com/remote/pkg/iops-2015-02-28
You can get the raw data from: https://www.dropbox.com/s/eknysmebhpwboa7/20150228-Seagate%208Tb%20Disk%20Testing.txt?dl=0

Seagate Archive 8TB Benchmark – Basic sequential tests

Quick basic testing on a RAID10 with 2MB chunk size, 1024 stripe width and 512 stride reveal they indeed move at rather leisurely speed.

Sequential write, 4MB block size: 344-354MB/s
Sequential read, 4MB block size: 376-386MB/s
hdparm buffered reads: 153-190MB/s.

But sequential, single thread speeds is not of consequence for us, but rather multi-thread random access. Seedboxes do severely thrash the storage – to the point most SSD caching is rather ineffective!

Seagate Archive 8TB Benchmark – Random access

First it is worth noting that best throughput was achieved with lowest thread count largest block size, maxing out at 311.2MiB/s for 8threads 64MiB block size – but also these drives are doing rather good relatively on higher thread counts, reaching 224.5MiB/s with 256 threads and 8MiB block size. But – the ultimate throughput is not of that high significance for our use case: Seedboxes.

Threads/Blocksize IOPS8 Threads16 Threads32 Threads48 Threads64 Threads128 Threads256 Threads512 Threads
512B236.5352.9245.3437.1437.1365267.2292.6
1 KiB265.9350.8406.7426.4432.6352.9276.9289.0
2 KiB279.5327.7356.8414.7393.5317267.9274.5
4 KiB247.1291.3296.7339.5299.9250.8232.3257.5
8 KiB236.1286.9297.8298.5316.5245.7240.0241.5
16 KiB225.7264.6271.6283.9275.6238.9234.1237.5
32 KiB212.9238.3273.8259262.2222.4218.2226.8
64 KiB197.0215.4224.8234.6208.6191.1194.4205.2
128 KiB160.172.9183.4188.6175.6160.8166.8173.2
256 KiB112.6130.2136.2141.1133.1123.9124.8124.3
512 KiB86.692.7101.697.393.594.589.292.6
1 MiB83.296.588.899.495.188.785.384.2
2 MiB58.864.267.869.167.465.965.067.9
4 MiB36.740.743.042.843.944.145.348.4
8 MiB22.123.923.424.8~23.625.227.230.3
16 MiB13.08.8
32 MiB9.5
64 MiB4.9

Interesting the performance keeps on increasing after certain thread count. Maybe firmware caching and queue management at work?

Let’s see how this array works against other machines:

Threads & BlocksizeSeagate Archive 4x8Tb RAID10Seagate/Toshiba 4x3Tb 7200RPM RAID52x2Tb 7200RPM HW RAID0160Gb SSD
8 Threads
16 MiB
138.416.9
32 Threads
512 KiB
101.698.436.4293.5
32 Threads
1 MiB
88.872.328.8186.7
64 Threads
512 KiB
94.588.335.3298.5
64 Threads
1 MiB
88.767.131.4274.0
64 Threads
2 MiB
65.943.221.8140.8
128 Threads
512 KiB
88.797.937.2303.4
256 Threads
256 KiB
124.8105.134.4583.4
256 Threads
512 KiB
89.2102.119.4308.2

Now you may say comparing RAID10 to RAID5 is not fair! Well – Actually, in both cases you get about 90-95% of the RAW underlaying storage speed, but we will also pit these 1 drive vs 1 drive below just for comparison.

The 2x2Tb HW Raid machine is very slow in comparison, hovering around 3.5-4.5 times slower than the 4x8Tb RAID10, and roughly 2-3 times slower than the 4x3Tb RAID5 comparison – But let’s remember it’s also half the disks (two vs four), so in reality disk per disk relative performance doesn’t have that huge difference, but it’s still very large and obviously there’s much room for improvement.

Seagate Archive 8Tb Benchmark – Single drive tests

These results are quite curious indeed, as they are rather hard to believe at first sight. BUT, that is the reason why we make these measurements and write everything down. Like Mythbusters put it: “Remember kids, the only difference between screwing around and science is writing it down.”

 Seagate ST8000AS0002Seagate ST3000DM001Toshiba DT01ACA300
64 Threads
4 KiB
100.76054.3
64 Threads
256 KiB
42.842.739.8
64 Threads
512 KiB
40.235.132.8
64 Threads
1 MiB
33.132.629.9
128 Threads
4 KiB
56.547.770.3
128 Threads
256 KiB
51.247.849.7
128 Threads
512 KiB
48.946.233.5
128 Threads
1 MiB
36.340.432.2

These are very surprising results indeed! The archive drives run at much lower RPM and they should not be in IOPS competitive against the 7,200RPM spinners. Something happening at the firmware level?

Then again, data does not lie. So at least with the 10secs per blocksize tests these 8Tb drives perform brilliantly fast. The read speeds are not that good to explain this. Further study is required.

Conclusions of the Seagate Archive 8Tb Benchmark

The Seagate’s ST8000AS0002 is performing amazingly well indeed! With the low power consumption, fast speeds and 5-year warranty we do sincerely wish Seagate is able to redeem itself with this drive model after the ST3000DM001 fiasko (Failure rates north of 40% annual). From business perspective, these drives makes a lot of sense and are dirt cheap in comparison to utilizing 3Tb drives with 2 or 3 year warranties.

Overally, if Seagate managed to get these drives to normal failure rates, that is somewhere around 5-10% annual, these will be very cost effective to operate in a datacenter environment compared to the current king of the market: Toshiba 3Tb DT01ACA300 drives.

Only real world, production use will show the true nature of these drives, but in the light of this data i know we are going to utilize quite a few of them in future, for now we will stick with a low quantity (couple dozen maybe) to feel them out and see how they keep functioning.

We’ve been eyeing towards 4Tb and 6Tb drives for a long time, but there is no cost-effective supplier here in Finland for them, so this 8Tb drive is a very welcome entrance to the market and we were quite surprised to see they are actually available for purchase in Finland. Thanks SystemaStore! For smooth operation it’s important to have a local supplier for smooth RMA (warranty replacement) procedures.

We will be sure to do more benchmarking on these drives shortly.

aleksi

One Comment

Leave a Reply

Your email address will not be published. Required fields are marked *