| Pages:
1
2
3
4
5
..
7 |
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Quote: Originally posted by coccigelus  | very interesting thread. Currently savvis the current data provider of Tradestation host their server at Equinix NY4 beside the main Tradestation
server. I did extensive research in the past about latency and I came across with a solution that let me reach TS server with 1 sub m/s through an
unmanaged dedicated server. The drawback of my solution however is that I have little control on the hardware I wish to have and not less important is
a very expensive solution. Peter's solution to colo is a very powerful idea. Steadfast is a top host service provider with a data Center at Cermack in
Chicago which is where IB server are. But since We use TS even if we trade through IB, be located to NJ or Chicago make little difference and
Steadfast is very expensive, way higher than 90$/months for 1u. When we use TS choose a location between Chicago or NJ make little if not at all
differnce but NJ is way cheaper than Cermack at Chicago.
Reading this thread I got an idea which I am not sure if Peter would be interested. I would like to create a group of ten traders and buy a 1/4 Shared
Rack (10u). The colocation for such configuration cost 400$ per month and the location is Secaucus Datacenter NY4 (Equinix). The hoster provider is a
well known hosting provider with super rating among the industries which I will provide to Peter if interested. The idea is to buy between 2/3
configuration of servers we think are good for our personal needs, give the funds to Peter and then buy the block of servers needed. Then pay in
advance 6 months for colo to Peter which mean 240$. Then in order to avoid newbies and later headaches for Peter state clear that If we do not want
to continue to host the server after the 6 months the server will be lost and that server will be available for a new user to be rent. In that way We
will have an initial cost of the server with exactly the hardware needed, a cost of only 40$/month for hosting, the best location possible which is
the same datacenter of TS and finally a true dedicated server.
|
There is merit to you idea. However I dont think its going to fly (yet) Most of the GSB users are buying servers for GSB, not execution. Hence Texas
is fine as we dont care on latency.
A rack full of GSB dual xeons is going to blow out power consumption wise.
Im paying $89 a month at steadfast. i will test the ping to TS. What ip did you use?
I am open to looking into the idea for my own trading, but last time I checked I thought I was 1ms away from TS at steadfast.
The big picture is IBLINK on my steadfast server is going to handle increasing amount of GSB traders execution. Its cost & time effect for small to
medium accounts. No data fees, no monitoring, no hardware costs. Just $14 r/t
|
|
|
coccigelus
Junior Member
Posts: 73
Registered: 11-7-2018
Member Is Offline
Mood: No Mood
|
|
Just for fun I just did a quick test with Steadfast/ my provider to IB/ Ts last node pingable, here the results:
Stead -) IB:
PING 208.245.107.3 (208.245.107.3) from 69.162.170.5 : 56(84) bytes of data.
64 bytes from 208.245.107.3: icmp_seq=1 ttl=248 time=2.77 ms
64 bytes from 208.245.107.3: icmp_seq=2 ttl=248 time=2.85 ms
64 bytes from 208.245.107.3: icmp_seq=3 ttl=248 time=2.89 ms
64 bytes from 208.245.107.3: icmp_seq=4 ttl=248 time=2.81 ms
64 bytes from 208.245.107.3: icmp_seq=5 ttl=248 time=2.78 ms
64 bytes from 208.245.107.3: icmp_seq=6 ttl=248 time=2.69 ms
64 bytes from 208.245.107.3: icmp_seq=7 ttl=248 time=2.68 ms
64 bytes from 208.245.107.3: icmp_seq=8 ttl=248 time=2.76 ms
64 bytes from 208.245.107.3: icmp_seq=9 ttl=248 time=2.71 ms
64 bytes from 208.245.107.3: icmp_seq=10 ttl=248 time=2.69 ms
Stead -) TS:
PING 4.35.20.209 (4.35.20.209) from 69.162.170.5 : 56(84) bytes of data.
64 bytes from 4.35.20.209: icmp_seq=1 ttl=62 time=1.45 ms
64 bytes from 4.35.20.209: icmp_seq=2 ttl=62 time=1.05 ms
64 bytes from 4.35.20.209: icmp_seq=3 ttl=62 time=1.08 ms
64 bytes from 4.35.20.209: icmp_seq=4 ttl=62 time=1.05 ms
64 bytes from 4.35.20.209: icmp_seq=5 ttl=62 time=1.13 ms
64 bytes from 4.35.20.209: icmp_seq=6 ttl=62 time=1.07 ms
64 bytes from 4.35.20.209: icmp_seq=7 ttl=62 time=1.07 ms
64 bytes from 4.35.20.209: icmp_seq=8 ttl=62 time=1.04 ms
64 bytes from 4.35.20.209: icmp_seq=9 ttl=62 time=1.08 ms
64 bytes from 4.35.20.209: icmp_seq=10 ttl=62 time=1.06 ms
--- 4.35.20.209 ping statistics ---
10 packets transmitted, 10 received, 0% packet loss, time 9013ms
rtt min/avg/max/mdev = 1.048/1.112/1.457/0.123 ms
X -) IB
PING 4.35.20.209 (4.35.20.209) from 69.162.170.5 : 56(84) bytes of data.
64 bytes from 4.35.20.209: icmp_seq=1 ttl=62 time=1.45 ms
64 bytes from 4.35.20.209: icmp_seq=2 ttl=62 time=1.05 ms
64 bytes from 4.35.20.209: icmp_seq=3 ttl=62 time=1.08 ms
64 bytes from 4.35.20.209: icmp_seq=4 ttl=62 time=1.05 ms
64 bytes from 4.35.20.209: icmp_seq=5 ttl=62 time=1.13 ms
64 bytes from 4.35.20.209: icmp_seq=6 ttl=62 time=1.07 ms
64 bytes from 4.35.20.209: icmp_seq=7 ttl=62 time=1.07 ms
64 bytes from 4.35.20.209: icmp_seq=8 ttl=62 time=1.04 ms
64 bytes from 4.35.20.209: icmp_seq=9 ttl=62 time=1.08 ms
64 bytes from 4.35.20.209: icmp_seq=10 ttl=62 time=1.06 ms
--- 4.35.20.209 ping statistics ---
10 packets transmitted, 10 received, 0% packet loss, time 9013ms
rtt min/avg/max/mdev = 1.048/1.112/1.457/0.123 ms
X -) Ts
PING 64.125.27.197 (64.125.27.197) 56(84) bytes of data.
64 bytes from 64.125.27.197: icmp_seq=1 ttl=59 time=1.23 ms
64 bytes from 64.125.27.197: icmp_seq=2 ttl=59 time=1.01 ms
64 bytes from 64.125.27.197: icmp_seq=3 ttl=59 time=1.51 ms
64 bytes from 64.125.27.197: icmp_seq=4 ttl=59 time=1.02 ms
64 bytes from 64.125.27.197: icmp_seq=5 ttl=59 time=0.983 ms
--- 64.125.27.197 ping statistics ---
5 packets transmitted, 5 received, 0% packet loss, time 4005ms
rtt min/avg/max/mdev = 0.983/1.152/1.513/0.204 ms
Imho both host provider are excellent. My point is that unless You know karl at Steadfast , Steadfast is ~ two times more expensive. I have used them
in the past although not directly.
I am personally interested mainly to a decent server for executions only but If I am going to buy a server I would pick up one that let me use it for
designing/testing as well. My biggest problem with a server for testing/developing located in the US is the massive lag due my location which is Asia
which is quite unpleasant and slow to handle.
I am executing as well at IB due the advantages of their brokerage and universal account. Out of curiosity, can You handle via IBlink adaptive orders?
If You are going to get a quarter/full rack and You have few empty available units available just let me know.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Its great you have done ping tests etc and chosen a data center well.
For me, I'm paying $89 at steadfast for whats $99 in your post, so I dont see the cost as an issue. Better value to get a rack like you say, but $89 a
month is a tiny part of my costs and not worth the effort to move a machine to another data center.
I normally last 3 years in a data center before Ive had enough of them. I have used steadfast for just under 7 years now. That says a lot.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
The post above on servers from theserverstore.com and datacenter Oplink has been updated with more details.
posted on 25-7-2018 at 12:33 AM
|
|
|
waldocktrades
Junior Member
Posts: 14
Registered: 12-7-2018
Member Is Offline
Mood: No Mood
|
|
processor/hardware
Has anyone tried the 2nd generation Threadrippers?
Currently debating between a Xeon 12 core @ 2.5ghz.
Thanks.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
I bought AMD Ryzen Threadripper 1950X & Gigabyte GA X399 AORUS MOTHERBOARD
A disaster. Much slower than intel. Sold it on ebay. The second hand dual xeon server is best ban for buck, followed by the I9 intel
|
|
|
edgetrader
Junior Member
Posts: 24
Registered: 16-5-2018
Member Is Offline
Mood: No Mood
|
|
The new Intel i9-9900K should come out this month, 16 threads for about 560 euros. Will have a soldered thermal interface material by default so no
need to delid. First overclocking results are 5.5GHz on all 8 cores. I guess 5GHz easy with air cooling and low voltage. Max memory is 64GB.
|
|
|
waldocktrades
Junior Member
Posts: 14
Registered: 12-7-2018
Member Is Offline
Mood: No Mood
|
|
Thanks for the input. I just had a system built and it took me two days to decide it wasn't nearly enough.
My assistant is going to get one fast hand me down...while I run workers on it.
|
|
|
Bruce
Member
Posts: 115
Registered: 22-7-2018
Location: Auckland - New Zealand
Member Is Offline
Mood: No Mood
|
|
New hardware recommendation,
What would provide the most kick for your dollar when it comes to CPU recommendation and RAM requirements? I'm going to get a new dedicated dev server
as I just upgraded my trade server. I've seen the previous discussions however prices move and newer CPUs are released which creates value
opportunities with the current CPUs. I appreciate the feedback. Thx.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Bank for buck, nothing is better than the dual xeon super micro from the server store.
The ram price is spectacular
|
|
|
cyrus68
Member
Posts: 171
Registered: 5-6-2017
Member Is Offline
Mood: No Mood
|
|
If you are using multiple datasets in GSB, you will run out of RAM faster than out of CPU capacity. My dual-socket Xeon, with 128 GB can handle it,
and more memory can be added, Moreover, it uses cheaper DDR3 RAM.
My i9, with 64 GB is maxed out and can only be upgraded to 128. It uses expensive DDR4 RAM. I agree with Peter that the Xeons are a better bet.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cyrus68 | If you are using multiple datasets in GSB, you will run out of RAM faster than out of CPU capacity. My dual-socket Xeon, with 128 GB can handle it,
and more memory can be added, Moreover, it uses cheaper DDR3 RAM.
My i9, with 64 GB is maxed out and can only be upgraded to 128. It uses expensive DDR4 RAM. I agree with Peter that the Xeons are a better bet.
|
The xeon is more the better option not because its a xeon, but because it has ddr3 ram (cheaper) I have a ddr4 xeon and the cost of ram upgrade is
horrific.
All 4 slots are full, so its got 64 gb only
|
|
|
cyrus68
Member
Posts: 171
Registered: 5-6-2017
Member Is Offline
Mood: No Mood
|
|
The cost of adding another 64 GB to my i9 is about $700.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
I'm likely to get another dual xeon server with 256 gb of ram. It will be for hire for $10 per 24 hours. Gives you a good feel if you want to invest
in some hardware yourself. Please email if this is of interest.
|
|
|
curt999
Junior Member
Posts: 51
Registered: 24-7-2017
Member Is Offline
Mood: No Mood
|
|
I just got a quad xeon poweredge r820 going to see how this fares with gsb..can get these barebones for about $450 with prcoessors no ram..192gb ram
is $300-350 for pc38500 ddr3..this has 32 cores with the e5-4620's x 4..need to load win10 workstation tho to get all processors seen
|
|
|
curt999
Junior Member
Posts: 51
Registered: 24-7-2017
Member Is Offline
Mood: No Mood
|
|
cluster HPC and GSB
im experitmenting with running GSB on a blade server cluster c7000 with 512 cores will post my results here after i get it setup..in general gsb
should also run on a multipurpose hpc cluster if setup correctly with individual servers using a network switch..the scalability for workers would be
huge esp if running some massive hardware like 4 c7000 blade server enclosures clustered together to offer over 2000 cores of processing power
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by curt999 | | im experitmenting with running GSB on a blade server cluster c7000 with 512 cores will post my results here after i get it setup..in general gsb
should also run on a multipurpose hpc cluster if setup correctly with individual servers using a network switch..the scalability for workers would be
huge esp if running some massive hardware like 4 c7000 blade server enclosures clustered together to offer over 2000 cores of processing power
|
Im fascinated how well this is going to work. Keep us informed.
|
|
|
avatartrader
Junior Member
Posts: 60
Registered: 1-10-2018
Member Is Offline
Mood: No Mood
|
|
I wanted to see if I could get some feedback or advice from those of you who have been testing GSB across different hardware platforms:
Currently, I am running my workers on a pair of dual-Xeon X5550 quad core servers. Admittedly, these are obviously a bit dated, but I happened to have
them on hand and so I figured that I would at least get started with them. So far, even with running a total of 20 workers across both of them, I am
only able to average about 5000-6000 systems/second, with full CPU utilization and about 50% memory utilization.
For my purposes, that's simply not going to do, so I was looking at getting some new hardware and on serversupply.com, I found quad 8-core E5-4650
servers for a very reasonable price (using the older DDR-3, so the RAM is also reasonable).
Does anyone have any experience on what type of performance I could expect to obtain with that generation of hardware and number of cores? Or does
someone have a lead on a better configuration at potentially a better price?
As much as I would love to get my hands on the type of hardware that curt999 is using, I'll need to be a bit more modest for now... 
Also, just out of curiosity, what is the pricing model of GSB cloud? I know that licenses come with 1 cloud worker, but I don't see a way to subscribe
to more.
Thanks!
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by avatartrader | I wanted to see if I could get some feedback or advice from those of you who have been testing GSB across different hardware platforms:
Currently, I am running my workers on a pair of dual-Xeon X5550 quad core servers. Admittedly, these are obviously a bit dated, but I happened to have
them on hand and so I figured that I would at least get started with them. So far, even with running a total of 20 workers across both of them, I am
only able to average about 5000-6000 systems/second, with full CPU utilization and about 50% memory utilization.
For my purposes, that's simply not going to do, so I was looking at getting some new hardware and on serversupply.com, I found quad 8-core E5-4650
servers for a very reasonable price (using the older DDR-3, so the RAM is also reasonable).
Does anyone have any experience on what type of performance I could expect to obtain with that generation of hardware and number of cores? Or does
someone have a lead on a better configuration at potentially a better price?
As much as I would love to get my hands on the type of hardware that curt999 is using, I'll need to be a bit more modest for now...
Also, just out of curiosity, what is the pricing model of GSB cloud? I know that licenses come with 1 cloud worker, but I don't see a way to subscribe
to more.
Thanks! |
A dual xeon2980 v2 - x 2690v2 is best bang for buck
refer here for rough guide of speed for dual cpu
https://www.cpubenchmark.net/multi_cpu.html
a dual 268x is about the speed of a i9
You dont get one cloud worker, you get a % of whatever is free.
Anyone can donate to the free cloud, and there is about 5 older servers and on occasions when im not using them- there are some of my dual xeons x
26xx
I havnt used the e5. Watch the power consumption . I wouldnt like the power bill of a quad xeon.
Im selling the user of a dual 2680 or 90 with 192 gb of ram for $10 a day. I could do a part or more that that subject to availability.
|
|
|
avatartrader
Junior Member
Posts: 60
Registered: 1-10-2018
Member Is Offline
Mood: No Mood
|
|
Right now, I am doing a run across my 2 dual quad core Xeon X5550's, my new i9 notebook w/ 32GB of RAM, and I am connected to a cloud worker on a Xeon
E5-2690 and the E5-2690 v2 is killing them all.
What is interesting is that on the Passmark benchmarks, a dual CPU Xeon E5-2680 v2 is roughly the same as the quad CPU Xeon E5-4650 I was looking at.
As I was writing this, a worker on an E5-2620 connected and is only doing about 500/min.
So, you're right, for the money, Xeon E5-2690 v2 seems to be the best choice.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by avatartrader | Right now, I am doing a run across my 2 dual quad core Xeon X5550's, my new i9 notebook w/ 32GB of RAM, and I am connected to a cloud worker on a Xeon
E5-2690 and the E5-2690 v2 is killing them all.
What is interesting is that on the Passmark benchmarks, a dual CPU Xeon E5-2680 v2 is roughly the same as the quad CPU Xeon E5-4650 I was looking at.
As I was writing this, a worker on an E5-2620 connected and is only doing about 500/min.
So, you're right, for the money, Xeon E5-2690 v2 seems to be the best choice.
|
You need to be careful here to compare apples with apples
If thats my 2690 (likely) then its running 1 to 3 workers only. (in the last few hours)
If you had say 8 workers, the collective speed will increase, but individual speed will drop.
Due to random seed in gsb, the identical machine with identical test could also vary a mild amount
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by avatartrader | Right now, I am doing a run across my 2 dual quad core Xeon X5550's, my new i9 notebook w/ 32GB of RAM, and I am connected to a cloud worker on a Xeon
E5-2690 and the E5-2690 v2 is killing them all.
What is interesting is that on the Passmark benchmarks, a dual CPU Xeon E5-2680 v2 is roughly the same as the quad CPU Xeon E5-4650 I was looking at.
As I was writing this, a worker on an E5-2620 connected and is only doing about 500/min.
So, you're right, for the money, Xeon E5-2690 v2 seems to be the best choice.
|
e2680 is now fully loaded close to 100% cpu.
you can check your speed now.
SOmeone is doing 27,28,29 min bars. Not the best compared to 29 30 31 min bars. (12,13,14 bars in day) compared to 14,14,14 bars in a day
|
|
|
avatartrader
Junior Member
Posts: 60
Registered: 1-10-2018
Member Is Offline
Mood: No Mood
|
|
I had paused and was doing a WF at that time, but just before I had paused my run, I had a couple of the E5-2690 workers drop down to an average of
about 600-700/min, so this may have been about that time, but I'm not sure how many workers were running in relation to cores.
It would be good to see what the average performance/min is with the optimal number of workers/core for a dual proc Xeon E5-2690.
I was doing 29,30,31 min bars, and at my peak last night I had 4 workers on the E5. For most of the time, they were doing between 1100 and 1400/min
each.
If they could handle about 600/min per worker and 1 worker per core, then I would estimate that they should be able to put out 12000/min.
When I fully load my X5550, I get about 300/min/worker on average, so I would hope that the E5 can do at least double that, if not more - not to
mention handle more workers thanks to the additional cores (even with a < 1:1 ratio). The benchmark for the dual E5-2690 v2 vs dual X5550 is close to
3x.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by avatartrader | I had paused and was doing a WF at that time, but just before I had paused my run, I had a couple of the E5-2690 workers drop down to an average of
about 600-700/min, so this may have been about that time, but I'm not sure how many workers were running in relation to cores.
It would be good to see what the average performance/min is with the optimal number of workers/core for a dual proc Xeon E5-2690.
I was doing 29,30,31 min bars, and at my peak last night I had 4 workers on the E5. For most of the time, they were doing between 1100 and 1400/min
each.
If they could handle about 600/min per worker and 1 worker per core, then I would estimate that they should be able to put out 12000/min.
When I fully load my X5550, I get about 300/min/worker on average, so I would hope that the E5 can do at least double that, if not more - not to
mention handle more workers thanks to the additional cores (even with a < 1:1 ratio). The benchmark for the dual E5-2690 v2 vs dual X5550 is close to
3x.
|
The gsbram runs the optimal number automatically. It maxes out till a cpu or ram limit is hit. 8 is typical.
cpubenchmark.net is a good guide. (apart from amd. Amd is woeful on GSB.) A dual cpu gives about 1.5 times the speed of a single. Unless you have full
control of the e5, you cant tell how much % of the machine you are running on. GSB purchasers have GSB admin mode, which tells you the entire machine
collective speed. However the limitation is if other workers that arent yours are on the machine, you cant see that.
|
|
|
engtraderfx
Junior Member
Posts: 98
Registered: 15-10-2018
Member Is Offline
Mood: No Mood
|
|
For interest, I found a few 2nd hand workstations, ex oil & Gas projects CAD machines around $1.2 to 1.5K? In perth, not sure if will post yet.
HP Z820 WorkStation - Dual Xeon 2.6 / Quadro 5000/ 256GB ECC RAM $1.5K
Bit of overkill with ram?
HP Z800 CAD WORKSTATION DUAL XEON 24 THREADS 96GB QUADRO 5000 $1.2K, multiple available.
Not many sites advertise specifically for trading computers, would computer's spec's for CAD be suitable, I notice my work also uses Z820 type
machines for heavy duty work.
|
|
|
| Pages:
1
2
3
4
5
..
7 |
![[*]](images/davis/default_icon.gif){kind=icon}
