Overclock.net › Forums › Specialty Builds › Servers › decreasing single execution time, which route to go....
New Posts  All Forums:Forum Nav:

decreasing single execution time, which route to go....

post #1 of 7
Thread Starter 
I have a server I use for daytrading. I'm trying to decrease my execution time by 50 milliseconds and I'm not sure which route to go, as I'm really not that computer savvy. Just enough to have a conversation and understand, but not enough to know what to do on my own. Currently using an older Dell poweredge 1950 w/ 16gb ram and PCI SSD. I'm trying to figure out what will give me the most performance, because I really have no idea if spending $10K on a server will decrease my latency 1 microsend or 1000 milliseconds, especially on a single execution/computation. I know in heavy load the newer server will outshine, but on single execution im not as sure. Same for ram upgrade, its ddr2, not sure if upgrading to DDR3 or 5 will yield the performance I'm seeking. My ssd's are ocz revo's from about a year ago. NIC is a 1g, but my understanding is upgrading to 10g will not yield anything because the server farm is bottlenecked to 1g above me in the chain.

Thanks for any help, hope I gave enough info to be helpful. For clarification, my questions basically break down to:

Will upgrading my server potentially decrease my SINGLE execution time, in a windows based server, enough to be in milliseconds?
Same for RAM?
Same for any other idea? Bios optimization, windows opt, etc???

Server has to be OEM spec as its colocated somewhere else. Cheers!
post #2 of 7
1950s are quite old, I wouldn't consider a 1950 with dual quads to be a High performance machine, especially with crappy ddr2 667mhz
Higher clock speeds on the cpu, higher Instructions per clock and a move to ddr3 will make a large difference in latency demanding applications.
How big is the application you are running? consider installing it in a 'Ram disk'.
3770/980ti
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 3770k @ 4.5GHZ Asus Z77M-pro EVGA 980Ti SC ACX 8GB 1600MHz 
Hard DriveHard DriveCoolingOS
256GB Samsung OEM SSD 128Gb Sandisk SSD Hyper 212 Evo Windows 10 
MonitorKeyboardMouseMouse Pad
Dell P2715Q 4K Apple Kbd Logitech G402 Steelseries QCK 
Audio
Senheiser CX215 
  hide details  
Reply
3770/980ti
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 3770k @ 4.5GHZ Asus Z77M-pro EVGA 980Ti SC ACX 8GB 1600MHz 
Hard DriveHard DriveCoolingOS
256GB Samsung OEM SSD 128Gb Sandisk SSD Hyper 212 Evo Windows 10 
MonitorKeyboardMouseMouse Pad
Dell P2715Q 4K Apple Kbd Logitech G402 Steelseries QCK 
Audio
Senheiser CX215 
  hide details  
Reply
post #3 of 7
Thread Starter 
Yes the 1950 is a dog for sure. Thing is my company wasnt looking to upgrade the entire server without some guidance from myself on what will be the increase in performance since its on their dollar. Do you have any idea what upgrading would do in terms on milliseconds per calc? I looked at some benchmarks and there are processors 10-20x faster now, pretty laughable how bad the 1950 is now. The ram disk looks awesome, learn something every day, the processes we are running could easily fit on a decent ram setup.
post #4 of 7
Quote:
Originally Posted by daytrayd View Post

Yes the 1950 is a dog for sure. Thing is my company wasnt looking to upgrade the entire server without some guidance from myself on what will be the increase in performance since its on their dollar. Do you have any idea what upgrading would do in terms on milliseconds per calc? I looked at some benchmarks and there are processors 10-20x faster now, pretty laughable how bad the 1950 is now. The ram disk looks awesome, learn something every day, the processes we are running could easily fit on a decent ram setup.

I wouldn't know the benefit in miliseconds, but im sure that even a $1000 custom built desktop would destroy the 6 year old cpus that are in your 1950.
Also, a newer NIC would help your latency on the network.

Before you buy a new server, I suggest you try a ram disk
3770/980ti
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 3770k @ 4.5GHZ Asus Z77M-pro EVGA 980Ti SC ACX 8GB 1600MHz 
Hard DriveHard DriveCoolingOS
256GB Samsung OEM SSD 128Gb Sandisk SSD Hyper 212 Evo Windows 10 
MonitorKeyboardMouseMouse Pad
Dell P2715Q 4K Apple Kbd Logitech G402 Steelseries QCK 
Audio
Senheiser CX215 
  hide details  
Reply
3770/980ti
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 3770k @ 4.5GHZ Asus Z77M-pro EVGA 980Ti SC ACX 8GB 1600MHz 
Hard DriveHard DriveCoolingOS
256GB Samsung OEM SSD 128Gb Sandisk SSD Hyper 212 Evo Windows 10 
MonitorKeyboardMouseMouse Pad
Dell P2715Q 4K Apple Kbd Logitech G402 Steelseries QCK 
Audio
Senheiser CX215 
  hide details  
Reply
post #5 of 7
What exactly is being calculated locally? Isn't day trading done over the internet? Perhaps a lower latency net connection would make a difference, not sure anything can be done local Pc wise. I mean I don't think there is anything to be done because that isn't the bottle neck. Unless I'm missing something about day trading?
 
VM Server
(17 items)
 
 
CPUGraphicsRAMHard Drive
Intel Ivy Bridge Core i7-3630QM nVidia GeForce GTX 680M 16GB DDR3 1600MHz Dual Channel Memory (2 SODIMMS) Hard Drive: Serial-ATA II 3GB/s 
Hard DriveOSMonitorPower
Hard Drive: Serial-ATA II 3GB/s Windows 10 Pro x64 17.3" FHD 16:9 (1920x1080) Battery: Smart Li-ion Battery (8-Cell) 
Audio
Sound Blaster Compatible 3D Audio 
CPUMotherboardGraphicsRAM
Intel Core i7 860 Biostar T5 XE Radeon HD 5870 Corsair 16GB  
Hard DriveHard DriveOptical DriveOS
Western Digital hard drive wd1001fals-00e8b0 Maxtor 300GB I don't need no stinking optical drive Microsoft Windows 7 Ultimate x64 
MonitorMonitorKeyboardPower
HP ZR24w 24'' Samsung SyncMaster 24" logitech wireless k360 Seventeam ST-850ZAF 850W ATX 
CaseMouseAudioAudio
Thermaltake V9 Black Edition Logitech G500 Programmable Gaming Mouse FiiO E7 USB DAC and Portable Headphone Amplifier Sennheiser HD555 Professional Headphones 
  hide details  
Reply
 
VM Server
(17 items)
 
 
CPUGraphicsRAMHard Drive
Intel Ivy Bridge Core i7-3630QM nVidia GeForce GTX 680M 16GB DDR3 1600MHz Dual Channel Memory (2 SODIMMS) Hard Drive: Serial-ATA II 3GB/s 
Hard DriveOSMonitorPower
Hard Drive: Serial-ATA II 3GB/s Windows 10 Pro x64 17.3" FHD 16:9 (1920x1080) Battery: Smart Li-ion Battery (8-Cell) 
Audio
Sound Blaster Compatible 3D Audio 
CPUMotherboardGraphicsRAM
Intel Core i7 860 Biostar T5 XE Radeon HD 5870 Corsair 16GB  
Hard DriveHard DriveOptical DriveOS
Western Digital hard drive wd1001fals-00e8b0 Maxtor 300GB I don't need no stinking optical drive Microsoft Windows 7 Ultimate x64 
MonitorMonitorKeyboardPower
HP ZR24w 24'' Samsung SyncMaster 24" logitech wireless k360 Seventeam ST-850ZAF 850W ATX 
CaseMouseAudioAudio
Thermaltake V9 Black Edition Logitech G500 Programmable Gaming Mouse FiiO E7 USB DAC and Portable Headphone Amplifier Sennheiser HD555 Professional Headphones 
  hide details  
Reply
post #6 of 7
Thread Starter 
Quote:
Originally Posted by subassy View Post

What exactly is being calculated locally? Isn't day trading done over the internet? Perhaps a lower latency net connection would make a difference, not sure anything can be done local Pc wise. I mean I don't think there is anything to be done because that isn't the bottle neck. Unless I'm missing something about day trading?

Yeah its over the internet, and that is probably our biggest latency externally speaking, but that is not what I'm trying to address. We are colocated in NY with our server farm at our data provider. My understanding is the data comes in at 10g, then is parsed throughout all the other internal chains at the warehouse at 1g, and then reaches our rack at 1g, where I am located. So upgrading the NIC to 10g might not provide much help. Maybe just the newer tech of the card will help. The problem I am having however is not external, but internally on our rack some newer servers are beating us by 50 milliseconds more or less. To provide a generic example of what we are doing, we are comparing probably 3-5 market data variables, including stock price, % change on day, and some other proprietary information, and if all are true then execute an order to purchase or sell. There are 2 pieces of software running to make this happen, basically a scanning tool and an execution platform. So its not a single calculation per say, but a single execution which requires the platforms working together. I am confused because of the internal difference in performance, which I am trying to track down to hardware vs software upgrades. The other server runs a newer software platform which is quicker, so that makes this problem harder to compare apples to apples. But I need to be able to beat those servers on performance. Since everything is on the companies money, I have little say what can be upgraded without some proof of what would help. If it were up to me I'd spend $10K on a awesome server and be done with it, but if that doesnt make up the 50 ms then I'd have to modify the software.
post #7 of 7
Oh, so a professional day trader smile.gif I thought you meant at home with scott trade or something rolleyes.gif

I don't think I know enough to answer this. I guess you could disable some Windows services, update drivers, clear temp files...? It's hard to imagine what 1 microsecond actually is, conceptually. It would be pretty funny if you made up that time because your server wasn't running the help-and-support service smile.gif (Okay I don't know if your server literally has that particular service running but you get the idea).
 
VM Server
(17 items)
 
 
CPUGraphicsRAMHard Drive
Intel Ivy Bridge Core i7-3630QM nVidia GeForce GTX 680M 16GB DDR3 1600MHz Dual Channel Memory (2 SODIMMS) Hard Drive: Serial-ATA II 3GB/s 
Hard DriveOSMonitorPower
Hard Drive: Serial-ATA II 3GB/s Windows 10 Pro x64 17.3" FHD 16:9 (1920x1080) Battery: Smart Li-ion Battery (8-Cell) 
Audio
Sound Blaster Compatible 3D Audio 
CPUMotherboardGraphicsRAM
Intel Core i7 860 Biostar T5 XE Radeon HD 5870 Corsair 16GB  
Hard DriveHard DriveOptical DriveOS
Western Digital hard drive wd1001fals-00e8b0 Maxtor 300GB I don't need no stinking optical drive Microsoft Windows 7 Ultimate x64 
MonitorMonitorKeyboardPower
HP ZR24w 24'' Samsung SyncMaster 24" logitech wireless k360 Seventeam ST-850ZAF 850W ATX 
CaseMouseAudioAudio
Thermaltake V9 Black Edition Logitech G500 Programmable Gaming Mouse FiiO E7 USB DAC and Portable Headphone Amplifier Sennheiser HD555 Professional Headphones 
  hide details  
Reply
 
VM Server
(17 items)
 
 
CPUGraphicsRAMHard Drive
Intel Ivy Bridge Core i7-3630QM nVidia GeForce GTX 680M 16GB DDR3 1600MHz Dual Channel Memory (2 SODIMMS) Hard Drive: Serial-ATA II 3GB/s 
Hard DriveOSMonitorPower
Hard Drive: Serial-ATA II 3GB/s Windows 10 Pro x64 17.3" FHD 16:9 (1920x1080) Battery: Smart Li-ion Battery (8-Cell) 
Audio
Sound Blaster Compatible 3D Audio 
CPUMotherboardGraphicsRAM
Intel Core i7 860 Biostar T5 XE Radeon HD 5870 Corsair 16GB  
Hard DriveHard DriveOptical DriveOS
Western Digital hard drive wd1001fals-00e8b0 Maxtor 300GB I don't need no stinking optical drive Microsoft Windows 7 Ultimate x64 
MonitorMonitorKeyboardPower
HP ZR24w 24'' Samsung SyncMaster 24" logitech wireless k360 Seventeam ST-850ZAF 850W ATX 
CaseMouseAudioAudio
Thermaltake V9 Black Edition Logitech G500 Programmable Gaming Mouse FiiO E7 USB DAC and Portable Headphone Amplifier Sennheiser HD555 Professional Headphones 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Servers
Overclock.net › Forums › Specialty Builds › Servers › decreasing single execution time, which route to go....