-
day 1 – Thunderbolt 2 speed tests with Maxx Digital 8 bay Thunderbolt 2 drive array
Hi –
we just got the new Mac Pro. I only had access to the horrible 3TB Toshiba SATA drives for these tests (so we really need to use HGST Ultrastar drives for more accuracy). But this is a 24 TB (RAID 5 configuration) eight 3TB drives, all Thunderbolt 2. Drive array is the Maxx Digital ThunderRaid (Thunderbolt 2) 8 bay. I don’t know if they will rename the box now.AJA System Test write speed is 900 MB/sec, and READ speed (using a 16 gig file, and 2K format) is 1100 MB/sec. Now, this is an exaduration (no different than Promise showed such great response on the R6 Pegasus Tbolt 1 when they first released it with the RAID 0 tests). It takes about 8 – 10 seconds to get up to 1100 MB/sec. It does not instantly show super fast speeds (as ALL Thunderbolt drive arrays perform) – it takes a second or two to get up to 600 MB/sec, and grows to 800 – 900 MB/sec in about 5 – 6 seconds, and then peaks at just under 1100 MB/sec. The performance is as good, or better than a Mini SAS 8 bay array with an ATTO R680 or Areca 1882x card using an “old” Mac Pro and the x16 lane slot (slot 1 or slot 2) – we can typically only get about 900 MB/sec with an 8 bay configuration, so the Thunderbolt 2 performance is impressive – but I am nervous about the “ramp up” in speed, because when you hit that PLAY button on your NLE software, you don’t want to make an excuse for any possible stuttering for the first few seconds.
I DO NOT have access to a Mac Book Pro with Thunderbolt 2 on it, but I am now dying to get one, so I can do REAL Thunderbolt 2 network speed tests. As you may recall, my tests with Thunderbolt 1 networking were dismal, and disappointing, and I am hoping that Thunderbolt 2 point to point will be a completely different set of results. But like I said – this is day 1.
Bob Zelin
Bob Zelin
Rescue 1, Inc.
ma*****@****rr.com -
4 Replies
-
Been running similar tests all week with a six core nMP directly attached to a Promise Pegasus2 RB 24TB RAID.
Also had a i7 MBP with Thunderbolt2 connection to the nMP through TCP/IP over Thunderbolt Bridge to the Pegasus2 RAID mounting it on the MBP desktop.
Both the nMP and MBP are also attached to an Areca 8GB/s FC Xsan through Atto 8GB/s ThunderLlink boxes but of course throughput of Xsan is limited to 800 MB/s from Thunderbolt1 link on Atto boxes.
With the nMP in DAS mode to the Pegasus2 R8 I’m getting 1123 MB/s in WRITE speed and 928 READ speed using the AJA DiskWhack Test set to 2K and 16 GB file size.
With the MBP in Thunderbolt2 Bridge mode to the nMP/Pegasus2 I’m getting 808 MB/s WRITE speed and 363 MB/s READ speed.
Attached is a screen grab showing both BMD and AJA speed tests of the nMP in DAS mode to the Pegasus2 R8:
Attached is a screen grab showing both BMD and AJA speed tests of the MBP in Thunderbolt2 Bridge mode to the nMP/Pegasus2 R8:
Here’s some photos taken with iPad showing set up:
However, using the same tool that Apple uses to test drive performance I was actually getting over 1200 MB/s WRITE speed and 1300 READ speed from the Pegasus2 in RAID0 DAS mode, so obviously the different disk speed tests are measuring different things.
The Promise support guys have published an interesting KB on their website that explains the disk caching optimization we went through to achieve the best results:
https://kb.promise.com/KnowledgebaseArticle10394.aspx
Individual disk speed is but one factor in overall sustained throughput using Thunderbolt 2 architecture … number of spindles, efficiency of RAID controller’s read-ahead cache and latency in TCP/IP over Tbolt2 bridging are all factors to take into account in overall system performance. Saturating the Tbolt2 pipeline requires that everything be firing on max throughput – 6 core and 12 core nMPs are great for CPU intensive tasks but maximizing Tbolt2 data I/O needs careful tweaking of the entire pipeline.
We’ll be demonstrating the new Mac Pros in DAS and Xsan mode at the Larry Jordan FCP X event on January 14th in Burbank and also at our X Pro monthly meeting on Saturday Jan 18th on The Lot in West Hollywood. We’ll also be presenting nMP, FCP X 10.1 and Xsan 3 at the January LACPUG meeting.
I’ve put together a short presentation that covers the different storage options now available using the nMPs in DAS, NAS and SAN configurations – details on our website:
https://www.lumaforge.com/index.html
Any questions, let me know or come along to one of the events.
Cheers and all the best for 2014.
NeilNeil Smith
CEO
LumaForge LLC
faster storage faster finishing
323-850-3550
http://www.lumaforge.com -
Thanks Bob, those speeds sound promising. Any idea when the TB2 version of the ThunderRAID will be available?
——————-
Los Angeles Cameraman
Canon C300 (x2), Zeiss CP.2 lenses, P2 Varicam, etc.
http://www.lacameraman.com -
we are using a Beta firmware version of the drive array. Hopefully very very soon – but I don’t have a firm answer for you yet.
And we have NOT started testing the expansion chassis yet (because the Tbolt 2 versions are STILL not ready) – we will do testing with all the existing RAID array and host adaptor cards (ATTO, Areca, etc.) to see if there are performance differences. Doing testing with a Tbolt 1 chassis (which we will do anyway) will probably be pointless, as the performance will not be as good.And we will be doing Thunderbolt 2 network testing as well (we are promised a Tbolt2 MacBook Pro in a few days, but we are VERY anxious to see if we can do a small workgroup with shared storage, without having to rely on 10GbE networking – I have no idea what the real life performance of this will be yet).
Again – this is day 1. And I am sure that Maxx Digital and Promise will not be the “only kid’s on the block” to be releasing Thunderbolt 2 products – I am sure that NAB2014 will be nothing but Thunderbolt 2 products !
Bob Zelin
Bob Zelin
Rescue 1, Inc.
maxavid@cfl.rr.com -
[Bob Zelin] “And we have NOT started testing the expansion chassis yet … we will do testing with all the existing RAID array and host adaptor cards (ATTO, Areca, etc.) to see if there are performance differences”
Bob, thank you for posting this. As an owner of both an ATTO and Areca Maxx Digital raid, I will be very interested in seeing those results posted.
Herb Sevush
Zebra Productions
—————————
nothin’ attached to nothin’
“Deciding the spine is the process of editing” F. Bieberkopf
Reply to this Discussion! Login or Sign Up